DISCUSSION PAPER SERIES May 2014 WolffHendrik Low Emission Zones and Adoption of Green Vehicles Keep Your Clunker in the Suburb: IZA DP No. 8180 of Labor Institute for the Study zur Zukunft der Arbeit Forschungsinstitut

Keep Your Clunker in the Suburb: Low Emission Zones and Adoption of Green Vehicles

Hendrik Wolff University of Washington and IZA

Discussion Paper No. 8180 May 2014

IZA

P.O. Box 7240 53072 Bonn Germany

Phone: +49-228-3894-0 Fax: +49-228-3894-180 E-mail: [email protected]

Any opinions expressed here are those of the author(s) and not those of IZA. Research published in this series may include views on policy, but the institute itself takes no institutional policy positions. The IZA research network is committed to the IZA Guiding Principles of Research Integrity.

The Institute for the Study of Labor (IZA) in Bonn is a local and virtual international research center and a place of communication between science, politics and business. IZA is an independent nonprofit organization supported by Deutsche Post Foundation. The center is associated with the University of Bonn and offers a stimulating research environment through its international network, workshops and conferences, data service, project support, research visits and doctoral program. IZA engages in (i) original and internationally competitive research in all fields of labor economics, (ii) development of policy concepts, and (iii) dissemination of research results and concepts to the interested public.

IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion. Citation of such a paper should account for its provisional character. A revised version may be available directly from the author. IZA Discussion Paper No. 8180 May 2014

ABSTRACT

Keep Your Clunker in the Suburb: Low Emission Zones and Adoption of Green Vehicles*

Spatial distribution and leakage effects are of great policy concern and increasingly discussed in the economics literature. Here we study Europe’s most aggressive recent air pollution regulation: Low Emission Zones are areas in which vehicular access is allowed only to vehicles that emit low levels of air pollutants. Using new administrative datasets from Germany, we assess the distribution of air pollution and the spatial substitution effects in green versus dirty vehicles. We find that LEZs decrease air pollution by around nine percent in urban traffic centers while pollution is unchanged in non-traffic areas. These results are driven by our finding that vehicle owners have an incentive to adopt cleaner technologies the closer they live to an LEZ. We reject the widespread concern that dirty vehicles contribute to higher pollution levels by increasingly driving longer routes outside of the LEZ. Back of the envelope calculations suggest that the health benefits of roughly two billion dollars have come at a cost of just over 1 billion dollars for upgrading the fleet of vehicles. Moreover, we find that non-attainment cities that decided not to include an LEZ but engaged in other methods (building ring roads, enhancing public transportation), experience no decrease in pollution.

JEL Classification: Q58, R48

Keywords: air pollution, low emission zones, PM10

Corresponding author:

Hendrik Wolff University of Washington Department of Economics 349 Savery Hall, Box 353330 Seattle, WA 98195-3330 USA E-mail: [email protected]

* We like to thank Maximilian Auffhammer, Maureen Cropper, Lucas Davis, Sumeet Gulati, Shanjun Li, Paulina Oliva and Elaina Rose for helpful discussions. The paper especially benefited from very detailed comments and suggestions by two anonymous referees and the editor. Thanks are also due to the seminar participants at the University of Washington, the University of California, Santa Barbara, and the AERE conference in Vancouver. We further like to thank the Umweltbundesamt of Germany to providing the air pollution data and the Kraftfahrtbundesamt of Germany for help discussing the vehicle emission categories. We are grateful for the research grant provided by the Royalty Research Fund of the University of Washington. All remaining errors are ours.

This paper is forthcoming in the Economic Journal. When citing the paper, please refer to the journal version, see DOI 10.1111/ecoj.12091. 1. Introduction

Recently, increased public health concerns have elevated the role of clean air policies. In

particular, focus is on PM10—the class of particulate matter smaller than 10 micrometers—a

major air pollutant from vehicle emissions. Because PM10 can enter the lungs and bloodstream, it is often considered the most lethal air pollutant. In the European Union alone, PM10 is estimated to cause 348,000 premature deaths annually. To put this into context, ozone—Europe’s second most deadly air pollutant—only causes about 21,000 premature deaths (Watkiss et al. 2005). In response to these health risks, the European Commission2 enacted the 2005 Clean Air

Directive, which marks an unprecedented attempt to mandate low levels of PM10. When cities violate the maximum allowable limits, mayors and local governments have to develop so-called clean air action plans. While these action plans can consist of various traffic measures, the most drastic has been the Low Emission Zone (LEZ), which defines an area where higher-polluting vehicles are completely banned from driving (Wolff and Perry 2010). In Germany, to deal with the large number of cities exceeding the EU threshold, the

government has categorized vehicles into four mutually exclusive classes of PM10 emissions. All 46 million German cars, buses and trucks are required to display a colored windshield sticker indicating its PM10 pollution class. As of 2010, 41 German cities have implemented LEZs banning vehicles based on the colors of these stickers. These zones have been controversial, however, because of the costs imposed on drivers and especially truck companies, for whom upgrading fleets to the appropriate sticker can be quite expensive.3 According to a recent online survey, over 91 percent of Germans disapprove of LEZs, considering them too bureaucratic with likely having little effect (DSM 2009). In an earlier survey, 70 percent of drivers stated they might drive around LEZs to avoid upgrading their vehicle (Vienken 2008). Despite these criticisms, LEZs have become a popular quick fix for local governments struggling to avoid the large financial penalties imposed for exceeding the EU limits. For example, a recently

2 The European Commission is the executive body of the European Union (EU). It proposes, implements, and enforces EU legislation for all of its twenty seven current member states. 3 Conversion to the next higher sticker costs 800 to 2,500 USD for passenger cars and 7,000 to 22,000 USD for larger vehicles and trucks, although conversion is technologically infeasible for some vehicles. Major newspapers’ headlines noted ‘Particulate Matter: The insanity of LEZs’ (Bild 2009), or ‘Driving Ban in LEZs: Much Dust for Nothing’ (Süddeutsche 2009). 2

announced penalty for the city of Leipzig is 700,000 Euro per day, or 1,050,000 U.S. Dollar (USD) per day, because of non-attainment with the EU clean air regulation. Germany is not alone in limiting vehicle use. Driving restrictions have been used for decades in some of the world’s most polluted cities. In 1989 Mexico City introduced the Hoy No Circula (HNC) policy which prohibits driving between 5am and 10pm one weekday per week based on the last digit of one’s license plate.4 Other forms of driving restrictions include partial and total bans (Italy, Athens, Amsterdam, Barcelona, and Tokyo); traffic cell architecture, such that vehicles can drive within cells but must take circumferential ring roads between cells (Goddard 1997, Vuchic 1999); traffic bans on days when air pollution exceeds certain thresholds (Milan and other Italian cities); and emissions fees combined with LEZ (in Greater London’s LEZ, larger vans and lorries pay a daily PM10 emission charge of 250 to 500 British pounds (392 USD 5 to 784 USD) if they do not meet the Euro IV PM10 standard ). Uncertainty about the effectiveness, however, creates difficulties to make informative decisions among policy options and to gain public support by policy makers. As a result, often choices seem ad hoc and regionally clustered.6 Despite the widespread use of driving restrictions, the related empirical literature is sparse. In a recent study, Davis (2008) analyzes the effect of Mexico City’s HNC policy on air quality. While he finds no change in weekday pollution levels, pollution actually increased on weekends and weekday late nights as drivers substituted towards driving when the HNC was not in effect. Davis shows this ineffectiveness is due to a surprising behavioral response: drivers circumvented

4 Similar license plate programs have been implemented in Athens (1982), Bogota (1998), Santiago (1986) and São Paolo (1997), San Jose (2005), La Paz (2003), all of Honduras (2008), and Beijing (2008). 5 In 2008, the Greater London Authority established one of the largest current LEZs in Europe, which roughly includes the area within the ring highway M25 that encircles Greater London. It restricts the most polluting vehicles according to the PM10 standard of the Euro IV norm, including buses, coaches, vans, utility vehicles, minibuses of weight 1.205 tones and more and diesel-engined heavy goods vehicles. This LEZ is different from London’s Congestion Charging Zone of eight pounds per day which operates on workdays during daytime only in London’s Center District (Leape 2006, TFL 2008). 6 Low Emission Zones (LEZs) are particularly popular in Europe, license plate programs implemented in Latin America and congestion charging mostly considered in northern Europe and major Asian urban centers. Instead, price-based policies that aim to limit congestion and emissions include road pricing and congestion fees. Singapore (1975), London (2003) and Stockholm (2006) charge fees to drive into the city center. While New York City's congestion fee stalled in the legislature, San Francisco is currently debating a six dollar fee to drive through downtown. Milan has combined congestion pricing and LEZs with its Ecopass program, which charges fees to drive downtown based on emissions-level. Despite the increase in price-based policies, command and control driving restrictions are still adopted. These latter policies are argued to be often easier to implement politically, technologically feasible, and relatively less expensive to enforce (Levinson and Shetty, 1992; Davis, 2008). 3

the restriction by buying older, more polluting second cars to have different license plates.7 Davis finds that the HNC is a high-cost solution—with social costs exceeding 300 million USD per year—given its negligible effect on air quality. While the counterproductive results in Mexico City were due to the particular design of the HNC,8 the German LEZ program may be more successful because it includes a differentiation by emission level, creating an incentive to adopt cleaner technologies.9 However, whether LEZs are effective is an empirical question. This study is related to the growing literature on estimating the costs of air pollution and its regulations (Dockery et al. 1993, Pope et al. 1995, Chay and Greenstone 2003 and 2005, Davis 2008). Our work adds to this literature presenting the first empirical paper on traffic restrictions which examines both the within-city and across city effects on pollution outcomes as well as the spatial substitution effects of the LEZ regulation on the adoption of abatement technology. To this end, the first task of this paper is to estimate the causal effect of LEZs on

PM10 levels using panel data of daily pollutions and weather conditions across Germany from 2005 through 2008. Both the pre-regulation pollution levels and the staggered nature of LEZ implementations produce rich identification for our estimation of the zones’ treatment effects.

One argument for the four-tier PM10 categorization is that it promotes a more rapid adoption of clean technologies since even vehicle owners who do not typically drive into an LEZ may want to keep the option value of free passage. To evaluate this, next this paper studies changes in the composition of the vehicle fleet. Using a unique administrative panel dataset of emission category and registration location of each vehicle from 2008 to 2010, we analyze the spatial substitution in vehicles’ emission categories attributable to LEZs.

We find that LEZs decrease PM10 around nine percent in traffic areas. The LEZ’s decrease in PM10 is larger for traffic stations inside the LEZs than those outside, although PM10 does not decrease at all in background areas away from major roads. This shows that cities target

pollution-reducing strategies at those traffic areas which are responsible for violating the PM10 limits. Recently several papers (Fowlie 2010, Auffhammer and Kellogg 2010) have shown that

7 Drivers also took more taxis, which were among the most polluting cars in Mexico when the HNC was enacted. 8 Meanwhile the HNC has been modified to include an exhaust monitoring program (Verifcación). Each car is affixed with a sticker indicating its class of emission and the cleanest cars are exempt from the HNC restrictions. 9 Small and Kazimi (1995) find heavy-duty diesel trucks have social costs per mile ten times higher than gasoline vehicles. Also Roson and Small (1998) find evidence that a small percentage of high-emission vehicles contribute the bulk of pollution and conclude that policies targeting dirty vehicles may be the best way to decrease emissions. 4

spatial (unintended) consequences can substantially change cost-benefit calculations of regulations on firms. In terms of individual choices, Graff Zivin and Neidell (2009) and Moretti and Neidell (2010) discuss settings in which the total welfare cost of air pollution is much larger due to avoidance behavior. LEZs may also cause unintended consequences. We reject the widespread concern that dirty vehicles contribute to higher pollution by increasingly driving longer routes outside of the LEZ. In terms of the spatial capital substitution, we find that drivers substantially increase the adoption of low emission vehicles the closer they live to an LEZ. In particular, the green commercially used vehicles, that presumably depend more on access to city centers, increased sharply by 88%. Privately used green vehicles increased by about 5%. Still, this represents a substantial shift in the spatial vehicle fleet composition due to the clean air regulation. Overall, back of the envelope calculations suggest that the health benefits of nearly two billion dollars have come at a cost of just over 1 billion dollars for upgrading the fleet of vehicles. Finally, this paper directly contributes to the EU policy debate of the design and choices of air pollution regulations. We find that all non-attainment cities that decided to not include an LEZ but engaged in other methods (building ring roads, enhancing public transportation), experience no statistical significant decrease in air pollution.

This papers proceeds as follows. Section two details the PM10 regulation and the implementation of LEZs. We describe our data in section three and discuss the empirical strategy

in section four. Section five presents econometric results of the causal impact of LEZs on PM10 levels and discusses the spatial substitution effects of high to low emission vehicles. Section six combines these results in cost-benefit analysis and we conclude in section seven.

2. Background 2.1 Air Pollution regulation in Europe 10 Motor vehicle emission is the primary source of ambient PM10 in urban areas, although, share of vehicle based PM10 can range widely both over time and spatially. We surveyed all

recent studies investigating the sources of PM10 in Europe (see Table A1 of Appendix A). The

10 Road transport is also largely responsible for all NOX, CO, benzene and black smoke emissions. While these toxins are regulated, threshold violations and health impacts are substantially higher for PM10. 5

main contributing factors are vehicle exhaust, resuspension of dust particles (caused by automobiles and by natural phenomena), combustion by industry and individuals, and other natural sources such as marine aerosol and pollen (Viana et al., 2008). In particular, for traffic

stations measuring PM10 in close proximity to roads, the share of vehicle exhaust is estimated to range from 25% to 55%. In contrast, for urban background stations the percentage of observed

PM10 that is attributed to vehicle emissions is only 8% to 23%. 11 12 In response to concerns about the health effects of PM10, the EU Clean Air Directives

introduced in 2005 EU-wide limits on ambient PM10 such that: (a) the daily average does not exceed 50μg/m³ on more than 35 days annually and (b) the yearly average does not exceed

40μg/m³. When any air pollution station exceeds the EU PM10 limit, the city is asked to develop a clean air action plan. In the beginning years of the EU regulation, often cities did not comply with the timely delivery of their action plans and the enforcement was lax. As a result, 70 percent of EU cities greater than population 250,000 had violated the limits at some point and, as shown in Table 1, the 35-day limit has caused violations in 81 German cities.13 As a consequence, in order to better enforce the legislation, the European Commission asked the European Court of Justice to impose financial penalties (Council Directive 2008/50/EC), such as the recently announced 700,000 Euro per day (1,000,500 USD per day) penalty on Leipzig for repeatedly violating the 35-day limit rule. The formula for the daily penalty payments is described in Wolff and Perry (2010). Further, in January 2009, the Europe Commission initiated infringement

proceedings against 10 EU countries that have not attained the PM10 limit. Moreover, EU citizens are entitled by law to demand action plans from local authorities.

Under Council Directive 1999/30/EC, a second phase of the PM10 policy was scheduled to begin on January 1, 2010. In this phase, the thresholds were to have been drastically tightened to a yearly average of 20µg/m³ and a maximum of seven days exceeding 50µg/m³. These limits would have been very difficult for many European cities to meet; for example, we estimate that

11 PM10 has long been linked to serious cardiopulmonary diseases, acute respiratory infection, trachea, bronchus and lung cancers (EPA 2004). Worldwide, about 6.4 million years of healthy life are lost due to long-term exposure to ambient PM10 (Cohen et al. 2005). 12 EU Clean Air Directives refers to a set of Council Directives including 1996/62/EC, 1999/30/EC and 2008/50/EC. 13 No German city violated the 40μg/m³ annual limit that did not also violate the exceedance day limit. 6

285 German cities would have violated the 2010 limits based on 2005-2008 emissions.14 In response, in 2008 the EU passed Council Directive 2008/50/EC, which abolished the second

phase of the PM10 policy, continuing the 2005 limits instead. While there is currently no indication that the 2010 limits will be reinstated, the prior threat of facing these limits were important in driving the widespread adoption of LEZs. 2.2 Low Emission Zones in Germany

Given the primacy of vehicle-based PM10, clean air action plans (AP) try to curtail emissions through (i) expanding public transportation (ii) utilizing ring roads (iii) improving traffic flow (iv) the implementation of an LEZ. The fourth option, implementing an LEZ, has emerged as the most drastic and controversial element of the action plans. The LEZs mostly cover city centers, but vary considerably in size. In Berlin, for example, the LEZ covers 88 square kilometers (km2), populated by 1.1 million people. Munich’s LEZ covers 44 km2 with 431,000 inhabitants and Frankfurt’s LEZ spans 110 km2. The largest LEZ in Stuttgart covers 207 km² with 590,000 inhabitants (see map of Figure B1 of Appendix B), while the nearby smaller LEZ in Illsfeld is only 2.5 km2 with 4,000 inhabitants. Figure 1 shows a map of current and planned LEZs and Appendix Table B1 and Table B2 list their characteristics. Each German vehicle—as well as each visiting foreigner—that wants to enter an LEZ must display a colored windshield sticker based on EU-wide emissions categories. There are four

PM10 classes for diesel vehicles. The highest emitting vehicles obtain no sticker (and hence cannot enter any LEZ), while red, yellow and green stickers are given to progressively “cleaner” vehicles, as shown in Table 2. There are two classes for gasoline vehicles, green and no sticker. In some cases vehicles can improve one class by retrofitting the engine or diesel particulate filter. The fine for illegally entering an LEZ is 40 Euros plus one driver’s license penalty point.15 There are exceptions that allow certain emergency and other work-related vehicles to enter LEZs

14 Even the national average in each year since 2005 violates both the 2010 exceedance day and annual average limits, as shown in Table 1. 15 There is a series of consequences for penalty points, ending with loss of driver’s license at 18 points. 7

without a sticker, including agricultural and forestry tractors; ambulances and doctor’s cars; vehicles driven by or carrying persons with serious mobility impairments; and police, fire brigades, Bundeswehr and NATO vehicles. The implementation date and the types of vehicles restricted by an LEZ vary across cities, as shown in Table 3. In Berlin, for example, all vehicles with a red sticker and “cleaner” (yellow and green) were allowed into the LEZ starting January 2008, while access has been further restricted to only green stickers since January 1, 2010. The LEZ of Dortmund (Brackler Strasse), on the other hand, has only permitted yellow and green sticker vehicles since January 2008. By 2012 more than 50 percent of all LEZ cities will allow only yellow sticker and cleaner vehicles, and by 2013 most of the LEZs will permit green sticker vehicles only. Of the 23 LEZs implemented in 2008, four began in January, eight in March, one in July and the rest in October. We categorize cities into various treatment groups based on this variation in implementation date and action plan components. Figure 2 illustrates the classifications. First, we divide stations

into 2 categories, ‘attainment cities’ (in Figure 2 abbreviated as AC) that do not violate the PM10 limit (and thus do not need to develop an action plan) and ‘non-attainment cities’ that must develop an action plan (AP). Next, we divide the non-attainment cities into ‘action plan only’ (APO) cities, whose action plan do not include an LEZ, and ‘LEZ cities’ (LEZ). Finally, we also separate out APO cities that include a 'future LEZ,' instituted in or after October 2008 (FLEZ).16

3. Data We collect a panel of air quality readings from January 2005 through October 2008 from the German Federal Environment Agency, the Umweltbundesamt (UBA). This data set includes a

combination of half-hourly, hourly or daily readings of PM10 for 554 stations in 388 cities. Stations are characterized by the UBA as being traffic stations, located on main arterial roads, or background stations, usually located in more residential and green areas such as public parks or at soccer fields. Using the coordinates for each station, we classify all stations as ‘inside’ or ‘outside’ of an LEZ.

16 There are some cities that have discussed implementing an LEZ but have not finalized a plan for doing so. We do not include these as FUTURE LEZ cities since a) it is unclear how serious these cities are about implementing an LEZ and b) these cities have PM10 levels closer to APO cities than FLEZ cities. 8

Figure 3 displays how PM10 levels have evolved since 2005. The way in which PM10 levels drastically vary over time over the range from below 20μg/m³ to over 100μg/m³ underscores the

difficulty modeling PM10 data; some of the variation is dependent on local weather conditions such as temperature, wind speed, rain and mixing layer height (Klinger and Sahn, 2008). To control for these factors, we collect the most detailed possible weather data available from the national weather service, Deutscher Wetterdienst. We obtain hourly weather readings for 34 stations and daily reading for 74 stations. Because the air quality and weather monitoring stations are not in the same location, we use geographic coordinates to match each air quality station with

the closest weather station. We only use the PM10 readings from stations that have a weather station within 50 kilometers in distance and 300 meters in altitude. The primary weather

variables are summarized in Table 4. After calculating daily weather and PM10 readings while 17 handling missing values , we end up with complete PM10 and matched weather data for 185 stations covering 122 cities.

Moreover, PM10 levels can highly depend on locational and temporal events. PM10 levels in our data often rise suddenly by several hundred percent, which could be attributable to activities such as open coal-fired barbeques or construction sites. While we cannot collect information on all particular events, we do aim to control for temporal changes as carefully as possible. First, we include as covariates all information on state-level specific school vacation and legal holidays, both obtained from Johannsen (2009). Second, we exclude New Years Eve and Day to avoid outliers caused by fireworks. We further control for the day of the week and we include flexible state-specific weather models. As long as confounding events are correlated with these variables

17 To calculate PM10 daily averages, we first linearly impute the missing hourly readings throughout the day. Once we have daily averages, we interpolate the missing daily averages for the 1.4 percent of days with no readings. Among those stations reporting half-hourly and hourly data, less than seven percent of days are missing observations for some hours, with over 70 percent of these being three hours or less. To make sure our results are not driven by changes in monitoring station composition, we restrict our analysis to stations that have “complete readings” for all of the years included in each analysis. We define a station as having “complete data” for 2005 to 2007 if there is data for at least 340 of the 365 days of the year. Since we only have data through October of 2008, a station has “complete data” for 2008 if there is data for 280 of the 305 possible days. Finally, on Google Maps, we studied the locations of all our stations in LEZ cities. This analysis led us to drop one station in Mannheim and one station in Berlin that were at rail yards, as both of these readings are presumably mostly picking up train emissions. 9

and uncorrelated with the LEZ treatment, our results should be unaffected. Finally, we include city-level 2006 population data from the Federal Statistical Office Germany Genesis database.18

4. Empirical Identification Strategy 4.1 Difference-in-Differences Approach To study the effect of LEZs on air quality we use difference-in-differences (see Meyer 1995; Bertrand et al. 2004) in which we compare LEZ cities to a set of control cities. This approach

calculates the difference between how much PM10 changes after adoption of LEZs in LEZ cities

and how much PM10 changes over the same time frame in control cities. This allows us to control

both for underlying differences between LEZ and control cities and temporal changes in PM10 levels common across all cities. We estimate Equation (1), where k indexes city, i indexes station and t indexes time

(1) ln ( , , ) = + + + , + ΨΧ , , + , ,

Our 푦main푘 푖 푡 parameter훼 훽1푝표푠푡퐿퐸푍 of interest푡 , 훽2,퐿퐸푍 measures푘 훽3 the푡푟푒푎푡퐿퐸푍 percentage푘 푡 by which푘 푖 푡 the휐푘 푖LEZ푡 affects PM10. The dependent variable, , , , is the훽3 average daily PM10 reading for each station. is an indicator variable for whether푦푘 푖 푡 a city has an LEZ and is an indicator for time퐿퐸푍 periods푘 after implementation of an action plan. , =푝표푠푡퐿퐸푍푡 is an indicator variable that equals one for non-attainment cities푇푟푒푎푡퐿퐸푍 after implementing푘 푡 퐿퐸푍 푘an∗ LEZ.푝표푠푡퐿퐸푍 , 푡, includes station-, city- 19 and time-specific covariates including weather variables, school vacation,Χ푘 푖 푡 holiday and day of the week indicator variables. Because the specific locational conditions of air quality stations

have a large impact on pollution readings, , , includes station fixed-effects in all models and

we analyze background and traffic stationsΧ푘 푖separately.푡 , , also include year-month fixed effects to control for any time trends or any other climatic effectΧ푘 푖 푡 that our weather model does not

18 In the German Genesis population file, the variable of population per city contains a number of missing observations in particular for small cities. We collected the missing population estimates by various internet searches. 19 Weather variables include daily values of mean temperature, mean temperature squared, maximum daily temperature, minimum daily temperature, 1-day lag mean temperature and maximum temperature, mean relative humidity, mean relative humidity squared, 1-day lag mean relative humidity, mean wind velocity interacted with whether it rained that day, maximum daily wind velocity, 1-day lag mean wind velocity, mean visibility, total precipitation, total precipitation squared, days without precipitation, mean temperature interacted with total precipitation, mean temperature interacted with mean relative humidity, mean temperature interacted with mean wind velocity, mean air pressure, 1-day lag mean air pressure. For regressions spanning multiple states, weather variables are interacted by state to control for the variation in climate across Germany. 10

capture. Identification comes from the assumption that, after controlling for changes in these

observables, PM10 levels would have evolved in the same way in treatment and control cities in the absence of an LEZ. Finally, in all analyses, we cluster standard errors by city to correct for serial correlation over time as well as spatial correlation across stations within a city (Bertrand et al. 2004).20

Clearly LEZs may not be the only driver of the observed changes in PM10. Local governments in non-attainment areas can choose including other measures in their clean air

action plans. To investigate the mechanisms driving PM10 reductions, first we compare PM10 in all AP cities to those of the attainment control (AC) cities. Next, to differentiate the effects of action plans with and without LEZs, we test the treatment of having an action plan only (APO) and having a future LEZ (FLEZ). We use Equation 1 to estimate these three treatments, replacing LEZ with AP, APO and FLEZ, respectively. In our analyses, we follow two different identification strategies to estimate these treatment effects. The first relies on matching cities based on pollution levels in the year 2005, and the second relies on matching cities based on location by comparing LEZ to FLEZ cities. These strategies are described next.

4.2 Matching Cities based on 2005 PM10

Our first identification relies on matching treatment and control cities based on similar PM10 levels prior to implementation of action plans. Specifically, we match cities on annual daily

averages of cities' highest-polluting station in 2005. Note, instead of using the PM10 average

across all stations, we use the cities’ highest PM10 polluting station because it is this station that

determines whether cities exceed the PM10 threshold. We match on 2005 because this is the last 21 year in which PM10 levels were not affected by action plans or LEZs. In order to obtain a group

of cities with similar initial conditions in PM10 levels, we use the 73 cities that have 2005

highest-station PM10 averages in the range of 25 to 35 µg/m³. We use the following rule for

selecting our range 25 to 35 µg/m³. First we calculated the 2005 median PM10 annual average

20 As a robustness test, we also have clustered by state, city-week and state-week and we found standard errors to be similar. Results are available upon request. 21 We acknowledge the limitation that we have no PM10 data prior to 2005 to detect potential pre-emptive behavior by cities to lower emissions. Note, however, below we find that all non-LEZ measures (i.e. increasing public transportation, building ring roads) do not lead to PM10 reductions between 2005 and 2008. This result does suggest that any potential pre-emptive non-LEZ measure was not successful in altering PM10 levels. 11

among highest-polluting-stations, which equals to 30 µg/m³, and then we add ±5 µg/m³ to get the range. We decided on the range of ±5 µg/m³ to obtain a mix of attainment, non-attainment, APO, LEZ and FLEZ cities. There is only one LEZ below this range and no attainment city above the range. Table B2 of Appendix B lists all cities and their treatment status. We assume that within this group of 73 cities, the 35 exceedance day threshold (none of these cities violated the yearly average PM10 standard) makes the designation of non-attainment status and subsequent development of action plans exogenous. Of these 73 cities, 22 cities serve as our control ‘attainment cities’ (AC) that have never violated the PM10 limits and do not have an action plan. Another twenty-two cities have an action plan but no LEZ (APO). We define these cities as APO cities if the first violation occurred in 2005 or in 2006. If the violation occurred instead in 2007 (or later), we drop the city from the analysis. This is necessary because not enough time has passed from the date of the violation until the end of our data (October 2008) in order to likely see an effect of the more long term action plan elements (i) to (iii). We also exclude 10 APO cities that developed an action plan 22 despite never violating the PM10 limit, as these are not unambiguously control or treatment cities. We ultimately use four cities that implemented an LEZ before October 2008 (LEZ) and seven cities with LEZs scheduled to begin between 2009 and 2011 (FLEZ). Table 5 compares

these different groups. Average 2005 PM10 levels are very similar across groups, ranging from 26.8 in attainment cities to 30.7 in APO cities. The cities only differ based on the number of exceedance days, with APO, FLEZ and LEZ cities exceeding the 50 µg/m³ threshold on more than 35 days, and the attainment cities less than 35 days. In this sense, this first identification approach can be also interpreted in the spirit of the regression discontinuity design (RDD) on the threshold of the number of exceedance days. This RDD advantage comes at a cost, however, since there are only four LEZ cities in our treatment group. The following second approach aims to increase the number of LEZ cities. 4.3 Geographical LEZ Approach

22 Some cities preemptively implement action plans to avoid violating the limits in the future, especially considering the tightened 2010 limits. There is one city, the city of Cologne, that implemented an LEZ despite never having violated the EU PM10 limit. Since our focus is on LEZ, we do not drop Cologne from our main regressions and, per suggestion of the referee, analyze Cologne separately below. 12

Our second identification strategy takes advantage of the staggered introduction of LEZs, individually comparing the earliest LEZ cities to nearby cities whose LEZ has not yet come into effect (FLEZ). There are multiple advantages of looking at each LEZ separately. First, weather and geography vary considerably across Germany and this allows us to fit a separate weather model for each region. Second, it ensures results aren't driven by other state or regional policies or events. Third, this geographical approach includes all cities with LEZs (compared to the first

approach that limited the analysis to cities only within the interval of 2005 PM10 levels from 25 to 35 µg/m³).23 Having more cities also allows us to analyze the heterogeneity between LEZs of different sizes. We make use of the staggered introduction of LEZs by comparing cities that instituted LEZs before October 2008 to other non-attainment cities that decided—for one reason or another—to introduce an LEZ at a later date. While this procedure comes at the cost of not

primarily matching on 2005 PM10 levels, we can instead match on the fact that (a) all cities plan to implement an LEZ, (b) geography and (c) city size. Identification in this section comes from

the assumption that there are no systematic differences in changes in LEZ cities’ PM10 levels based on when they implemented their LEZ beyond the effect of the LEZ. 4.4 Common Trends Assumption One concern with our differences-in-differences framework is that differential trends in the

level of the PM10 between treatment and control cities can make the identification strategy invalid. Furthermore, our above strategy to use FLEZ cities (as control units) requires that the

timing of the LEZ implementation is unrelated to the prior PM10 levels. To test for these common

trends, Panel a of Table 6 regresses 2007 PM10 levels on date of LEZ introduction, along with the station, time, holiday and weather covariates used in all other regressions below. In our main regressions, the PM10 levels are taken from the year 2007, which immediately precedes the introduction of LEZs in 2008. Panel 6a shows that the coefficients on LEZ start date are small and insignificant, for both traffic and background stations. Similarly (as per suggestion of the referees), Panel 6b to Panel 6c repeat the analysis by using the years 2005 and 2006 as well as

the changes in PM10 levels between years (Panel d). Overall, the regression results at traffic stations show strong support that the timing of the introduction of the LEZ is not influenced by previous pollution levels. Inconsistent with these results is column (3) of Table 6c, which shows

23 Out of the 12 cities that implemented LEZ by March of 2008 (see Appendix B), we cannot analyze Schwäbisch Gmünd, Ilsfeld or Dortmund (Brackeler Road) because these cities have insufficient data. 13

that cities with larger 2006 PM10 levels at background stations introduce LEZs later. This result is significant at the 10% level. Note that this effect of column (3) is in the opposite direction of the concern that high polluting cities introduce LEZs earlier—not later. To summarize, overall

Table 6 indicates that the prior PM10 levels of a city do not predict when the LEZ is introduced. This supports our identification strategy to use FLEZ cities as control units in our Differences-in- Differences framework. Finally, Figure 3 as well as the Figures in Appendix C depict average daily PM10 levels over the entire sample period for treatment and control cities utilizing each matching approach. These Figures visually illustrate that these groups of cities are not inherently different in terms of differential trends of PM10 levels and provides additional support to our identification strategies. 4.5 Spatial Substitution in Emission Categories of Vehicle Fleet This analysis examines whether LEZs promote spatial adoption of cleaner vehicles and technologies. We perform this analysis using an administrative panel dataset containing data on vehicle emission categories and registration location from the German Federal Motor Transport Authority (Kraftfahrtsbundesamt Flensburg). Total number of private and commercial vehicles by emission category are observed for all districts from 2008 to 2010. We estimate Equation (2),

(2) ∆=sicpαβ pc + pc Min{} Distanceij+γ pc X i + ε ipc

where Δsicp = sicp2010 - sicp20008 is the change in the share of vehicles with sticker color c = {green, yellow, red, no sticker} in county i of vehicle usage type p = {private, commercial}. This adoption function depends on the minimum distance in kilometers of the centroid of county i to the set of LEZ cities j and Xi includes characteristics of county i, county income per capita, population size and state fixed effects. For each sticker color and vehicle usage type the regression function (2) is estimated separately by OLS with robust standard errors resulting in

the set of estimates of interest βpc.

5. Results This section presents our results of estimating the effect of the clean air action plan policies

on air quality. The following two subsections 5.1 and 5.2 are based on the ‘matching on PM10 in 2005’ approach described in 4.2, while subsection 5.3 presents our results based on the geographical matching approach.

14

5.1 Effect of Non-Attainment Status One challenge with evaluating Germanys LEZs is to disentangle the ‘LEZ effect’ from the other possible clean air action plan instruments enacted simultaneously. To investigate into this,

first we test the overall effect of violating the PM10 standard by comparing cities that developed

any type of action plan (AP) to attainment cities (AC). Table 7 compares PM10 levels in 2005, the period before being found in non-attainment, to 2008, when cities violating the standard had at least two to three years to implement clean air action plans.24 Columns 1 and 2 show that APs

in general have not had a significant effect on PM10 at either stations located in high traffic areas or stations located in background areas. This apparent non-effectiveness of APs may be driven by the heterogeneity between plans with and without LEZs. To test this, we next isolate APs from those cities that have no LEZ planned (APOs) and those who have future LEZs (FLEZ). One may expect that the APO treatment effect may be greater than the FLEZ treatment, since FLEZ cities, anticipating the

planned LEZ, may not meanwhile take other (costly) steps to combat PM10. Columns 3 and 4 of

Table 7 show that APOs have had no significant effects on PM10. Next columns 5 and 6 show the FLEZ treatment effect. Again, there are no significant changes at either traffic or background stations. In summary, it does not appear that the APO measures of building ring roads, increasing

public transportation or enhancing the traffic flow have had any influence on PM10 levels. Moreover, these results imply that prior to implementing their LEZs, FLEZ cities take no other

effective measures to combat vehicle-based PM10. Thus in the regressions to follow, we feel

comfortable attributing changes in current LEZ cities' PM10 to the LEZ rather than other APO or FLEZ policies. 5.2 Effect of LEZs In this section, we isolate the treatment of having LEZs as part of an action plan as specified in Equation 1. Table 8 compares the LEZs that began before October 2008 to the attainment control cities. The timing of the difference-in-differences is to compare the months in 2008 following the implementation of the LEZs, to the same months of the previous year 2007. We

24 These regressions only include January through October since we do not have PM10 data for November and December 2008. 15

use April through October data, since the Mannheim, Reutlingen and Leonberg LEZs didn’t take effect until March 2008 and we are allowing a one month lag for cities to adjust to the LEZs.25

The main result of Table 8 is that LEZs on average over all cities have lowered PM10 levels by nine percent (columns 1) in urban traffic areas. This main result of this paper will be tested below by using alternative identification strategies below. To investigate the situation at background stations, columns 2 and 4 show statistically insignificant increases of four to seven percent. Thus the decrease in PM10 along major roads within the LEZ is not being realized

outside of these high-traffic areas, which shows that PM10 from road traffic is a local pollutant.

Note, no LEZ city violated the PM10 standard because a background station exceeded the EU threshold. Hence, cities first focus on reducing emissions in those traffic areas that caused the city to violate the standard. All of the background stations in the above LEZ cities are located outside of the LEZs. This also suggests that drivers of non-conforming vehicles did not increase driving around LEZ to avoid upgrading their vehicles, but we will investigate this question further below.26 One concern with our 2005 matching identification strategy is that results depend on our

fixed range of 25 to 35 µg/m³ of pre-intervention PM10 emissions. In Table 9 we symmetrically increase this range to investigate how sample selection affects our results. The upper panel is

based on matching on the 25-35 PM10 range, corresponding to Table 8. The panels below

symmetrically increase the PM10 ranges to [23, 37], [21-39], [20-40] and (0, +∞) µg/m³. As one might expect, widening the 2005 emission ranges attenuates regression results (hence showing reduced magnitudes and decreased significance) as the similarities in initial PM10 levels decreases for the group of included treated and control cities. However, overall, the results are fairly robust. We view the strictest 25 to 35 emission range as our preferred matching strategy used in the manuscript. Before proceeding, we note that our results are robust to alternative specifications. In particular whether or not to include weather covariates, holiday or population

25 In some of the early LEZs, like Berlin, drivers were often only given warnings and not tickets in the first few weeks after the LEZ was introduced (Climate Company, 2009). 26 We expand on the analysis of background stations in the next subsections. In Section 5.3 shows results of background stations that are located within an LEZ. In Section 5.4, we examine Berlin’s LEZ (in which there are both traffic and background stations inside and outside the LEZ) which allows for explicit analysis of the hypothesis whether PM10 increases around LEZs due to avoidance behavior. 16

information does not change our overall treatment effect. Details on these robustness checks are provided in Appendix D. To investigate the heterogeneity between LEZs, Table 10 displays the treatment effects from

regressing each LEZ city separately to the same set of control cities. At traffic stations, PM10 decreases in the range of five percent in Cologne27 to 13 percent in Mannheim. At background stations, again, none of the treatment effects are significantly different from zero. In summary,

while there is some heterogeneity, all the LEZs are associated with significant decreases in PM10 at traffic stations. 5.3 Geographical LEZ results This subsection presents the results from our geographical identification strategy, which compares LEZ cities to nearby future LEZ (FLEZ). First, Table 11 displays the estimates from combining all the LEZs and their control cities. Across all traffic stations, column 1a shows that

LEZs are associated with a 7.3 percent decrease in PM10 (which is qualitatively similar to the result of columns 1 and 3 of Table 8 in the previous section). Auffhammer, Bento and Lowe (2009) show that within U.S. counties in non-attainment of pollution standards, pollution abatement plans have bigger effects in the areas that cause the non-attainment than those that do

not violate the standards. To look for any similar heterogeneity, we analyze PM10 pollution only at cities’ dirtiest stations and find the treatment effect increases to minus 10.7 percent (column 2a). One main question with the implementation of LEZs is whether air pollution decreases inside LEZs only, or whether outer areas of cities also benefit from the adoption or cleaner vehicles. To

explore this question we coded the PM10 traffic measurement stations as inside or outside of the LEZ. Column 3a of Table 11 shows that the treatment effect at stations inside LEZs is 8.6 percent, slightly larger than the average treatment effect. In comparison, at traffic stations outside

of LEZs, PM10 decreases by a statistically insignificant 3.6 percent (column 4a). These results

imply that the benefits of PM10 reduction within the zones are not fully carried over to the traffic

27 It is an interesting to note, that Cologne realized the lowest PM10 reduction (5%). Cologne is the only LEZ city that implemented an LEZ despite never violating the EC regulation. In Cologne therefore not the police is enforcing the regulation, but the much less representative agency “Stadt Koeln”, which only issued a couple hundred of tickets. See http://www.spiegel.de/auto/aktuell/start-der-umweltzonen-kontrolle-vielleicht-strafe-spaeter-a- 526151.html and http://www.ksta.de/innenstadt/verkehr-kaum-bussgelder-in-der- umweltzone,15187556,21405348.html. 17

stations outside of LEZs, a result that we will investigate in more detail in the case of Berlin

below. Hence, again, we do not find statistical support for the stated hypotheses that PM10 levels increase around LEZs due to increased driving by dirty vehicles that cannot enter the LEZs. Again, consistent with the results of the previous subsection, at background stations LEZs are not associated with any significant change in PM10, as displayed in Columns (1b) to (4b). To further study the heterogeneity of LEZ cities, Figure 4 shows the treatment effect coefficients from comparing both background and traffic stations of each LEZ city to neighboring future LEZ cities.28 The cities in the figure are ranked in descending order by the number of inhabitants within the LEZ,29 such that Berlin has the most people residing within its

LEZ. Consistent with the above findings, PM10 decreases at all LEZ cities’ traffic stations (except for the smallest two cities by inhabitants, Ludwigsburg and Leonberg, where there is no

statistically significant change in PM10). At 12 percent, this decrease is greatest in Berlin—the most-populous LEZ—and the treatment effects tend to diminish with lower populous LEZs. The effect of LEZs is again more heterogeneous for background stations. There are significant

increases in PM10 for Stuttgart, Mannheim, Reutlingen and Tübingen, while the changes are insignificant for Berlin, Hannover, Cologne30 and Ludwigsburg.31 The above Differences-in-Differences estimates could overstate our LEZ treatment effect

if PM10 data are subject to mean reversion. In particular, this is a serious concern due to the noisy 32 outcome of PM10 which in reality depends on many variables that we cannot control for. If cities are assigned into nonattainment status purely due to random shocks of emissions and then

PM10 reverses to its mean, our numerical LEZ estimate will also represent this mean reversion

28 See Table E1 of Appendix E for the control cities used for each LEZ city. The numerical results of the regressions are provided in Table E2. 29 The number of inhabitants in the LEZ of Reutlingen and Tübingen has not been published. By geographical analysis of the boundaries of the LEZ (available from Climate Company, 2009), we estimate that the number inhabitants for Reutlingen and Tübingen is 78,523 and 78,300. 30 The statistically insignificant yet relatively large decrease at Cologne’s background station could be because one station is located about 340 meters southwest of a major interstate. 31 Compared to the city by city results of the ‘matching on 2005’ identification, these city by city geographical regressions use much fewer control cities and the standard errors are larger. In both versions we cluster standard errors by city to be conservative. 32 Local construction sites, open coal barbeques or similar events can temporary drastically increase PM10 emissions. If PM10 exceeds 50 mg per cubic meter more than 35 days per calendar year in part due to such outlier events, unrelated to traffic, and then PM10 reverses in the following years to its statistical mean, this phenomenon is described in the literature as mean reversion (Chay et al. 2005). 18

effect and hence overstate the true LEZ treatment effect. In following regressions of Table 12 and Table 13, we introduce placebo treatments as if the LEZs were introduced one year earlier. Hence using the year 2007 as the placebo treatment and 2006 as the control year in our Differences-in-Differences framework.33 We perform this placebo test for both

(a) the 2005 PM10 matching strategy and (b) the geographical matching approach.

In our placebo regressions below, we expect to see no significant effect of LEZs on PM10 because no LEZ was introduced prior to 2008. The two tables below list the results for each matching approach (a) and (b). The Tables are built analogue to the main original LEZ regression Table 8 and Table 11 presented earlier, with the LEZ treatment effect replaced by the placebo dummy. In summary, the general lack of significance of the placebo dummies in Table 12 and Table 13 indicates that both of our matching strategies are robust to these tests of mean reversion. 5.4 Spatial Spillover Effects: The LEZ of Berlin The LEZ policy of Berlin is of particular interest. It covers over 88 square kilometers and it is the largest LEZ in terms of the 1.1 million inhabitants that live within the LEZ. Furthermore, Berlin already tightened its regulation, such that only green sticker vehicles have been allowed to enter the zone since January 1st of 2010.34 As we are fortunate to have a particular large set of background and traffic stations both located within and outside of the LEZ of Berlin, we next analyze this city further.35 We first compare stations inside the LEZ of Berlin to the stations outside of the LEZ. In particular, the control group is defined as the set of four stations that are located outside of the LEZ of Berlin yet strictly within the greater urban area of Berlin. Column 1 of Table 14 shows that traffic stations within the LEZ experience a 6.0 percent decrease in PM10 relative to traffic

33 Flagging treatment and control cities identically to our main regressions in the paper where 2008 is the treatment year and 2007 is the control year. 34 This is a drastic tightening which implies that from 2010 onward, 62% of all commercially used vehicles in Germany (including all commercial trucks and buses) are banned to enter the city of Berlin since they do not have the green sticker. Further 13% of all privately used vehicles are non-green and hence excluded to enter the city of Berlin. See Table 15 for details. 35 This section uses fifteen air pollution stations. Four stations are located within the LEZ boundary of Berlin, four are located within the greater city of Berlin, but are outside of the boundary of the LEZ and seven stations are drawn from the FLEZ cities that serve as controls in below regression. Of these, eight are traffic stations and seven are background stations. 19

stations outside the LEZ. This decrease could be either because PM10 emissions decrease within the LEZ, or because emissions increase outside of the LEZ as vehicles are forced to drive around it. To explore this, column 3 separately compares Berlin’s four inside- and four outside-LEZ stations to the seven nearby control stations used in the geographical approach. Traffic stations within the LEZ experienced a significant reduction of 15.0 percent, which is the largest treatment effect among all our LEZ cities and likely attributable to the size of the LEZ and the stricter implementation scheme. Stations located outside of the Berlin LEZ also reduce PM10 levels by a significant 9.1 percent. Again, this is substantially larger than the (insignificant) average reduction of 3.6 percent for all German outside-LEZ traffic stations, as displayed in Table 11, column 4a. These results suggest that the benefits of adopting cleaner vehicles are also realized outside of the LEZ. In other words, even if more vehicles need to drive outside of the LEZ to circumvent it, this effect would be more than offset by the increased use of cleaner vehicles. Columns 2 and 4 show that background stations within the LEZ see no significant change in

PM10 relative to those outside of the LEZ, again supporting the evidence that adopting an LEZ does little to improve air quality in areas away from major roads. Finally, the case of Berlin also allows us to check our identification assumption whether FLEZ cities are appropriate control cities. Note that the 6% decrease in Column (1) is ‘reflected’ in the LEZ versus FLEZ regression of Column (3): the ‘inside LEZ’ coefficient of -.15 is exactly .06 larger compared to the ‘outside LEZ’ coefficient of minus 9%. This shows internal consistency of these two separate estimations in Columns (1) and (3). This lends additional support to our identification assumption of using FLEZs as credible control cites.36 5.5. Spatial Substitution between low and high emission vehicles

One important argument for the four-tier PM10 categorization is that it prompts a more rapid adoption of cleaner technologies, as even those who do not typically drive into an LEZ may want to keep the option value of free passage. To test this, next we construct an unique panel dataset of German vehicles to analyze spatial substitution effects in purchasing new vehicles and retrofitting existing vehicles attributable to LEZs.

36 Also note that the same calculations hold for the background stations showing that the two (insignificant) treatment effects of minus 0.046 and minus 0.040 in column (4) are roughly equal to the (insignificant) treatment effect of minus 0.007 in column (2). 20

Data from the German Federal Motor Transport Authority (Kraftfahrtsbundesamt Flensburg) includes yearly observations of the total number of private and commercial vehicles by emission category (green, yellow, red or no sticker) for all districts from 2008 to 2010, reported on January 1 of each year. Table 15 summarizes the composition of the vehicle fleet; with 84%, the vast majority of vehicles now belong to the green sticker category. The changes over the two year time period are drastic. While on average the private vehicle fleet increased by only 1.3%, the green sticker group increased over-proportionally by 5.2%. This increase was driven by a drastic reduction of the red and no sticker vehicles, which decreased by 28 and 23 percent respectively (panel (A) Table 15). For commercially owned vehicles these changes are even more remarkable. As displayed in panel (B), their green sticker category increased by 88 percent, while red and no sticker vehicles decrease by 21 and 26 percent respectively. Because commercial vehicles are used for business activities and often rely on access to the city center, the pressure to upgrade the commercial vehicle fleet is more pronounced. Figure 5 displays changes in the share of green sticker private vehicles by county between 2008 and 2009 as a function of the county’s distance to the next LEZ.37 The change in green sticker share is between 0.01 and 0.035 share points, while counties close to an LEZ experience the largest increase in green stickers. Visually, Regensburg and Bonn are outliers. It turns out that these cities’ special circumstances explain their greater adoption of green sticker cars. In 2007, the local government of Regensburg announced an LEZ for spring 2008, then decided to postpone the introduction until September 1, 2008. This date was then again postponed to be tentatively scheduled for January 2010 (Stadt Regensburg, 2008). To this date the LEZ of Regensburg is still not implemented (Climate Company, 2010). It is therefore likely that the inhabitants of Regensburg responded to these announcements by preemptively upgrading their vehicles. The second outlier, the city of Bonn with 300,000 inhabitants, is very well connected to the LEZ city of Cologne (one million inhabitants) via a system of highways with mostly no speed limits, providing an incentive for Bonn’s drivers to obtain green stickers Next, Figure 6 shows how the share of the private vehicles without stickers (i.e. the highest- emitting vehicles) changed as a function of distance to the closest LEZ. All counties experience a decrease in the share of dirtiest vehicles, while once again the counties closest to an LEZ see the

37 We use February 15, 2009 to determine the status of the city whether it contains an LEZ or not. 21

largest drop. Similarly, Figure 7 and Figure 8 show the change in shares of yellow and red sticker vehicles, respectively. Here the changes are more uniform across different counties. This is not surprising because these middle emissions categories are banned by few of the current LEZs. Consistent with these figures, we present regression results in Table 16, which show that the adoption of green technology increases the closer the vehicle is registered to a city that has an LEZ. The change between 2008 and 2010 in each county’s percentage of vehicles with green stickers is regressed on the distance from the county to the nearest LEZ city. In Table 16, the results are given separately for private and commercial vehicles and repeated for the change in percent of vehicles without stickers, the dirtiest. In particular, we find that for each km closer a vehicle is to an LEZ, the incentive to upgrade to a green sticker increases by about one percent for commercial vehicles and 0.6 percent for private vehicles. Hence, again, the effect is larger for commercial vehicles, as these rely more on access to city centers for business. In summary, we find evidence that the introduction of LEZs creates an incentive for drivers to substitute towards lower-emitting vehicles. The closer a county is to an LEZ, the more likely its citizens have been to substitute away from the dirtiest cars and towards the cleanest cars. The incentives are particularly strong for the commercial vehicles, which more aggressively updated their fleet due to the LEZs.

6. Cost Benefit Analysis To get sense of LEZ’s efficiency, we use our results to calculate back-of-the-envelope costs and benefits of LEZs.38 In order to calculate the changes in health benefits, we use epidemiological estimates measuring the effect of PM10 on long-term mortality from Medina et al. (2004). For the calculation we apply the Value of the Statistical Life of $7.9 million (2008$) to monetize these benefits (EPA 2000). Using our city-specific estimates (from the geographical)

approach of Section 5 and applying the reductions in PM10 to the exposure of the number of inhabitants residing within each LEZ, we monetize these health benefits to be $1.98 billion dollar.

38 Appendix F provides the details about the specific calculations involved of the costs and the benefits . 22

These health benefits stand against the costs of the LEZ program. The largest costs are due to the upgrading of the vehicle fleet. To calculate these costs, we use the vehicle spatial substitution results of Section 5. First, we use the vehicle registration data to fit regressions of the change in share of green-sticker private and commercial vehicles39 on the distance from an LEZ. To avoid counting vehicles that would have switched to the green sticker category in the absence of the LEZ regulation, we use as the baseline the change in share of green stickers for the point furthest away from any LEZ. For each location, we subtract this baseline from our regressions’ predicted change in share of green sticker vehicles. This is what we consider as the change in share of green stickers attributable to the LEZ. We then multiply this coefficient by the location’s number of vehicles to obtain the number of new green vehicles attributable to the LEZ. Finally, we sum these numbers for all locations and multiply by the weighted average cost for upgrading cars, buses and trucks to get the total cost of $1.09 billion dollar due to LEZ induced vehicle upgrading. These calculations are clearly approximate in nature and we omit some potentially important factors. First, the benefits may even be larger if congestion decreased within the LEZ reducing the amount of the time spend in stop and go traffic. This time saving effect would need to be compared to the additional time needed for those drivers that need to driver longer routes to circumvent the LEZ. Second, ancillary pollutants are not considered in the calculation of the health benefits. Third, business within the LEZ and outside of the LEZ can adversely or positively be affected. Fourth, the upgrading of the vehicle can potentially have other benefits to the driver, such as having a safer or more comfortable vehicle. Fifth, in our calculation we only considered the benefits to the residents that strictly live within the LEZ area. In the Appendix F we include calculations for all inhabitants of the cities which increases the benefits from $1.98 to $5.22 billion. Sixth, we consider only the changes at traffic stations. In some cities, however, also background stations were affected. Taking these changes into account reduces the estimated health benefit by $0.3 billion, in the case of the geographical matching approach. For our ‘2005 matching’ approach, the welfare estimate however remains unchanged. With these limitations in mind our main results indicate health benefits of roughly two billion U.S. dollars, which came at a cost of about one billion U.S. dollar to upgrade the German vehicle fleet.

39 Since vehicle registration data is measured as of January 1of every year, the data really measures vehicles purchased or upgraded in 2007 versus the vehicles purchased or upgraded in 2008. 23

7. Conclusions With over half of the world’s population living in increasingly motorized cities, urban traffic policies are receiving much interest to manage congestion, protect public health and to reduce emissions of air pollutants. A wide range of tools have been implemented, including the license plate program, permanent driving bans, congestion pricing, traffic cell architecture, temporary driving bans, the building of ring roads or the implementation of Low Emission Zones (LEZs). Uncertainty about the effectiveness, however, creates difficulties to make informative decisions among policy options and to gain public support by policy makers. As a result, often choices seem ad hoc and regionally clustered. This paper is the first to study the effect of Low Emission Zones (LEZs), which is one of the most aggressive tools intended to rapidly decrease air pollutants, widely adopted across the European Union but also exists in Asia (i.e. Tokyo) and variations of the program are frequently discussed in combination with other traffic options, i.e. to exclude traffic charges for low emission vehicles in the U.S.40 and elsewhere. Our main findings are that (a) LEZs reduce particulate emissions by nine percent, whereas other air quality policies (which do not include a LEZ) had surprisingly no effect; (b) avoidance behavior of driving around the LEZ does not lead to significant spatial spillover effects; (c) there is heterogeneity across zones, with larger LEZs having stronger impacts; (d) spatial vehicle fleet composition changed drastically as a response to the announcements of LEZs in Germany. Overall, our back of the envelope calculations predict health benefits of nearly two billion dollars that have come at a cost of just over 1 billion dollars for vehicle upgrading. While many more cities will have to implement stricter policies soon to circumvent EU penalties, this is the first timely paper to assess this popular and rapidly growing policy. More studies of related policies (congestion charging, public transportation etc.) are in order to better inform this public policy debate and to evaluate the relative efficiencies41 of competing policies.

40 In New York Bloomberg’s popular plan was to introduce a fee system of congestion charges with exceptions and discounts for certain low-emission vehicles. Other major U.S. cities are discussing similar programs. 41 See Fowlie et al., (2011) for a recent study on the relative efficiency of policies effecting stationary and non- stationary pollution sources. 24

Literature Auffhammer, M. and Kellogg, R. (2011). ‘Clearing the Air? The Effects of Gasoline Content Regulation on Air Quality’ American Economic Review. vol. 101, pp. 2687-2722. Auffhammer, M., Bento, A. and Lowe, S. (2009). ‘Measuring the Effects of the Clean Air Act

Amendments on ambient PM10 concentrations: The Critical Importance of a Spatially Disaggregated Approach’ Journal of Environmental Economics and Management vol. 58(1), pp. 15-26. Bertrand, M., Duflo, E. and Mullainathan, S. (2004). ‘How Much Should We Trust Differences- In-Differences Estimates?’ Quarterly Journal of Economics vol. 119, pp. 249-275. Bild (2009): Feinstaub: Der Irrsinn mit den Umweltzonen. Article of the Bild Zeitung, published November 24th, 2009. Accessed online on Bild.de on November 24, 2009. Chay, K.Y. and Greenstone, M. (2003). ‘The Impact of Air Pollution on Infant Mortality: Evidence from Geographic Variation in Pollution Shocks Induced by a Recession’ Quarterly J. Economics vol. 118(3), pp. 1121–67. Chay, K.Y. and Greenstone, M. (2005) ‘Does Air Quality Matter? Evidence from the Housing Market’ J. Political Economy vol. 113(2), pp. 376–424. Chay, K.Y., McEwan, P.J. and Urquiola, M. (2005). ‘The Central Role of Noise in Evaluating Interventions That Use Test Scores to Rank Schools’ American Economic Review, vol. 95(4), pp. 1237-1258 Climate Company. (2009). GEMB Gesellschaft fur Emissionsmanagement und Beratung mbH, Geschaftsfuehrer Michael Kroehnert, HRB 101917 AG Berlin Charlottenburg, http://www.umwelt-plakette.de/ Cohen A.J., Anderson, H.R., Ostro B., Pandey, K.D., Krzyanowski, M., Kunzli, N., Gutschmidt, M., Pope, A. Romiau, I., Samet, J.I., and Smith, K. (2005).. ‘The global burden of disease due to outdoor air pollution’ J Toxicology and Environmental Health, PartA vol. 68, pp. 1– 7. Council Directive 08/50/EC Journal of the European Communities L 152 pp. 1–44. Council Directive 96/62/EC Journal of the European Communities 296, 21.11. pp. 55–63. Council Directive 99/30/EC Official Journal of the European Communities, L 163/60.

25

Davis, L. (2008). ‘The Effect of Driving Restrictions on Air Quality in Mexico City’ Journal of Political Economy vol. 116(1), pp. 38-81. Dockery, D., Pope, C., Xu, X., Spengler, J., Ware, J., Fay, M., Ferris, B., and Speizer, F. (1993) ‘An association between air pollution and mortality in six U.S. cities’ New England Journal of Medicine vol. 329(24) pp. 1753-1759. DSM (2009) Online Survey of Deutschlands schnellste Meinung. http://www.bild.de/BILD/dsm/deutschlands-schnellste-meinung.html (last accessed 24 November 2009). EPA. 2000. Guidelines for Preparing Economic Analyses, Office of the Administrator. EPA- 240-R-00-003, December 2000. EPA. 2004. Air quality criteria for particulate matter (Final Report, Oct 2004). U.S. Environmental Protection Agency, Washington, DC, EPA 600/P-99/002aF-bF. Fowlie, M. (2010). ‘Emissions Trading, Electricity Industry Restructuring, and Investment in Pollution Control’ American Economic Review, vol. 100(3), pp. 837-869. Fowlie, M, Knittel, C.R. and Wolfram, C. (2011). ‘Sacred Cars? Cost-Effective Regulation of Stationary and Non-stationary Pollution Sources.’ American Economic Journal: Economic Policy, 2012, 4(1), pp. 98-126. Goddard, H. C. (1997). ‘Using Tradeable Permits to Achieve Sustainability in the World’s Large Cities’ Environmental and Resource Economics vol. 10 (1) pp.63–99. Graff Zivin, J. and Neidell, M. (2009).‘Days of haze: Environmental information disclosure and intertemporal avoidance behavior’ Journal of Environmental Economics and Management, vol 58 (2), pp. 119-128. Johannsen (2009): Schulferien.org website maintained Moritz Johannsen, 16548 Glienicke, Germany.

Klinger, M. and Sahn, E. (2008). ‘Prediction of PM10 Concentration on the Basis of High Resolution Weather Forecasting.’ Meteorologische Zeitschrift vol. 17(3), pp. 263-272. Krewski D., Burnett R.T., Goldberg M.S., Hoover, K., Siemiatycki, J., Abrahamowicz, M., Villeneuve, P. and White, V. (2000). ‚Re-analysis of the Harvard Six cities Study and the American Cancer Society Study of air pollution and mortality’ Cambridge, MA: Health Effects Institute.

26

Kunzli N, Kaiser R., Medina S., Studnicka, M., Chanel, O., Henry, M., Horak Jr., F., Puybonnieux-Texier, V., Quenel, P., Schneider, J., Seethaler, R., Vergnaud, J.C., and Sommer, H.. (2000). ‘Public-health impact of outdoor and traffic-related air pollution: a European assessment’ Lancet vol. 356, pp.795–801. Leape, J. (2006). ‘The London Congestion Charge’ Journal of Economic Perspectives, vol 20(4), pp.157-176. Levinson, A. and Shetty, S. (1992). ‘Efficient Environmental Regulation: Case Studies of Urban Air Pollution in Los Angeles, Mexico City, Cubatao and Ankara.’ Policy Research Working Paper no. 942, World Bank, Washington, D.C. Medina S., Plasencia, A., Ballester, F., Mucke, H.G., and Schwartz, J. (2004). ‘Apheis: public

health impact of PM10 in 19 European cities’ Journal of Epidemiology and Community Health vol. 58, pp. 831-836. Meyer, B. (1995). ‘Natural and Quasi-Experiments in Economics.’ Journal of Business & Economic Statistics vol. 13, pp. 151-161. Moretti, E. and Neidell, M. (2011) ‘Pollution, Health, and Avoidance Behavior: Evidence from the Ports of Los Angeles’ J. Human Resources vol. 46(1), pp. 154-175. Pope III, C.A., Thun, M.J., Namboodiri, M.M., Dockery, D.W., Evans, J.S., Speizer, F.E., and HeathJr., C.W. (1995). ‘Particulate air pollution as a predictor of mortality in a prospective study of U.S. adults.’ American Journal of Respiratory Critical Care Medicine vol. 151(3), pp.669-74. Roson, R. and Small, K. eds. (1998). Environment and Transport in Economic Modeling. Kluwer Academic Publishers, Dordrecht, Boston and London. Seethaler, R. (1999) ‘Health Costs Due to Road Traffic-Related Air Pollution: An Integrated Assessment of Austria, France and Switzerland.’ Third WHO Ministerial Conference of Environment and Health, London. Small, K. and Kazimi, C. (1995). ‘On the Costs of Air Pollution From Motor Vehicles.’ Journal of Transport Economics and Policy vol. 29, pp.7-23. Stadt Regensburg (2008): Stadtradt Regensburg, Ausschuss fuer Stadtplanung, Umwelt, Verkehr und Wohnungsfragen. Beschluss vom 27. Februar, 2008, Regensburg.

27

Süddeutsche (2009): Fahrverbot in Umweltzonen: Viel Staub um Nichts. Article of the Süddeutsche Zeitung, published November 24th, 2009. Accessed online on sueddeutsche.de on November 24, 2009. Transportation for London. (2010). ‘TFL’s Report to the Mayor on the Low Emission Zone Variation Order Consultation’ LEZ VO Consultation Report, September 2010, http://www.tfl.gov.uk/assets/downloads/roadusers/lez/LEZ/lez-vo-report-to-mayor-sept- 2010.pdf U.S. Environmental Protection Agency. 2000. Guidelines for Preparing Economic Analyses, Office of the Administrator. EPA-240-R-00-003, December 2000. U.S. Environmental Protection Agency. The Benefits and Costs of the Clean Air Act, 1990-2010. Report to the U.S. Congress, November 1999. Van Zelm, R., Huijbregtsa, M.A.J., den Hollander, H.A., van Jaarsveldd, H.A., Sautere, F.J., Struijsb, J., Wijnenc, H.J. and van de Meent, D. (2008). ‘European characterization factors

for human health damage of PM10 and ozone in life cycle impact assessment.’ Atmospheric Environment vol. 42, pp.441–453. Viana, M., Kuhlbusch, T.A.J., Querol, X., Alastuey, A., Harrison, R.M., Hopke, P.K., Winiwarter, W., Vallius, M., Szidat, S. and Prevot, A.S.H. (2008). ‘Source apportionment of particulate matter in Europe: A review of methods and results.‘ Journal of Aerosol Science vol. 39, pp. 827–849. Vienken (2008): Summary of Survey about LEZ: Deutsche Autofahrer zweifeln am Sinn und Zweck von Umweltzonen: Knapp 70% wuerden nach Moeglichkeit Umweltzonen umfahren. Survey by DA Direkt Versicherung. Accessed online at www.presseportal.de/pm/17575/1165262/da_direkt_versicherung on November 24, 2009. Vuchic, V. R. (1999). Transportation for Livable Cities, CUPR Press. Watkiss, P. (2003). ‘The London Low Emission Zone Feasibility Report: A Summary of the Phase 2 Report to the London Low Emission Zone Steering Group.’ http://www.tfl.gov.uk/assets/downloads/roadusers/lez/phase-2-feasibility-summary.pdf Watkiss, P. S., Pye S. , Holland, M. (2005). CAFE CBA: baseline analysis 2000 to 2020. Report to the European Commission DG Environment, Brussels.

28

Wolff, H. and Perry, L. (2010) ‘ Trends in Clean Air Legislation in Europe: Particulate Matter and Low Emission Zones’ Review of Environmental Economics and Policy. Vol. 4(2) pp.293-308.

29

Table 1: EU PM10 limits

Panel A: European Union PM10 pollution thresholds Phase 1 Phase 2 (now defunct) originally planned to since 1 January 2005 start 1 January 2010 Yearly average limit 40 µg/m³ 20 µg/m³ Daily average (24-hour) limit 50 µg/m³ 50 µg/m³ Allowed number of exceedences per year 35 7 Numbers of German cities violating the 81 285* standard

Panel B: Germany violations of PM10 limits 2005 2006 2007 2008 24.4 26.2 23.1 21.2 National average PM [µg/m³] 10 (5.2) (5.5) (5.3) (4.9) 19.6 26.8 16.2 11.6 Mean number of days** above 50 µg/m³ (20.9) (21.1) (15.8) (12.9) Cities in violation of 2005 standard 36 65 31 18 Cities in violation of 2010 standard 226 246 200 134 *The calculation of the expected number of cities violating the 2010 standard is based on the number of cities that would have violated the standard between 2005 and 2008 either because of exceedance days or high annual averages **: Average of the highest exceeding station per city; Standard deviations in parentheses

Table 2: German vehicle stickers Sticker categories No sticker Red Yellow Green

Euro 2 or Euro 3 or Requirement for diesel Euro 1 or Euro 4 or Euro 3 Euro 1 with Euro 2 with vehicles worse with particle figure particle filter particle filter Without 3- Euro 1 with Requirement of gasoline way catalytic regulated catalytic vehicles converter converter or better

30

Table 3: German LEZ restrictions 2008 to 2012

31

Table 4: Summary of weather data Weather Variables Unit Mean St. Dev. Min Max Daily average temperature 1C 9.6 7.7 -27 31 Daily min temperature 1C 5.6 6.9 -29.7 24.4 Daily max temperature 1C 13.8 8.9 -23.1 40.2 Daily avg. vapor pressure 1 hpa 9.9 4.1 0.2 26.9 Daily average air pressure 1 hpa 981 49 679 1047 Daily avg. relative humidity % 78.1 12.9 7 101 Daily avg. wind speed 1 m/s 2.6 1.1 0 10 Daily max wind speed 1 m/s 10.9 4.9 1.3 64.8 Daily avg. cloud cover Tenths 7.1 1.3 0 9 Sun in day 1 hour 4.8 4.4 0 16.7 Precipitation during day 1 mm 2.1 4.7 0 158 Snow depth cm 4.1 28.6 0 550

Table 5: Treatment and control characteristics Number of 2005 highest- Avg. number of cities polluting station avg. exceedance days Attainment cities (AC) 22 26.8 22.4 Action Plan only cities (APO) 22 30.7 36.8 AP with LEZ after Oct. 2008 (FLEZ) 7 30.0 34.7 AP with LEZ before Oct. 2008 (LEZ) 4 28.9 28.9 These are the cities that make up our sample for the 2005 PM10 matching analysis.

32

Table 6: Effect of LEZ start date on 2007 PM10 levels

Panel 6a: Effect of LEZ start date on 2007 PM10 levels All Stations Traffic stations Background stations

(1) (2) (3) LEZ start date -0.000173 -0.000225 0.0000811 [0.000958] [0.00117] [0.000890] Observations 24942 15478 9464 Adjusted R-squared 0.719 0.711 0.681

Panel 6b: Effect of LEZ start date on 2005 PM10 levels All Stations Traffic stations Background stations (1) (2) (3) LEZ start date 0.00137 0.00164 0.00167 [0.00163] [0.00112] [0.00262] Observations 19409 10757 8652 Adjusted R-squared 0.694 0.697 0.622

Panel 6c: Effect of LEZ start date on 2006 PM10 levels

All Stations Traffic stations Background stations (1) (2) (3) LEZ start date 0.00282 0.00251 0.00386* [0.00193] [0.00199] [0.00205] Observations 25813 16359 9454 Adjusted R-squared 0.717 0.718 0.672

Panel 6d: Effect of LEZ start date on change in PM10 levels between 2006 & 2007 All Stations Traffic stations Background stations

(1) (2) (3) LEZ start date -0.00120 -0.00188 0.0000886 [0.00118] [0.00139] [0.00231] Observations 23085 13991 9094 Adjusted R-squared 0.469 0.478 0.467 All regressions include year-month fixed effects, weather, holiday, station type and population covariates. Robust standard errors in brackets are clustered by city, *** p<0.01, ** p<0.05, * p<0.1

33

Table 7: Effect of Action Plans on Log PM10

Matching based on 2005 PM10 in range 25 to 35 Action Plans Without Action Plans With Future All Action Plans (AP) LEZs (APO) LEZs (FLEZ) Traffic Background Traffic Background Traffic Background Stations Stations Stations Stations Stations Stations (1) (2) (3) (4) (5) (6) AP treatment 0.0117 0.0404 -0.0125 0.0440 0.0316 -0.0523 [0.0366] [0.0466] [0.0357] [0.0484] [0.0530] [0.0706] Observations 28859 21236 22378 16380 12746 11532 Adj. R-squared 0.657 0.618 0.656 0.622 0.604 0.608 All regressions include year-month fixed effects, weather, holiday, station type and population covariates. Regressions include data for January-October 2005 vs. 2008. Robust standard errors in brackets are clustered by city, *** p<0.01, ** p<0.05, * p<0.1.

Table 8: LEZ vs. Attainment cities

Matching based on 2005 PM10 in range 25 to 35 All cities Cities > 100,000 Traffic stations Background stations Traffic stations Background stations (1) (2) (3) (4) LEZ treatment -0.0910*** 0.00724 -0.0686* 0.0448 [0.0241] [0.0285] [0.0302] [0.0354] Observations 6723 7704 2896 4280 Adj. R-squared 0.657 0.591 0.653 0.653 All regressions include year-month fixed effects, weather, holiday, station type and population covariates. Regressions include data for April-October 2007 vs. 2008. Robust standard errors in brackets are clustered by city, *** p<0.01, ** p<0.05, * p<0.1.

34

Table 9: LEZ vs. Attainment cities: Sample Selection Effects Matching based on 2005 PM10 in range 25 to 35 All cities Cities > 100,000 Traffic stations Background stations Traffic stations Background stations (1) (2) (3) (4) LEZ treatment -0.0910*** 0.00724 -0.0686* 0.0448 [0.0241] [0.0285] [0.0302] [0.0354] Observations 6723 7704 2896 4280 Adj. R-squared 0.657 0.591 0.653 0.653 Matching based on 2005 PM10 in range 23 to 37 All cities Cities > 100,000 Traffic stations Background stations Traffic stations Background stations (1) (2) (3) (4) LEZ treatment -0.0822*** 0.00973 -0.0686* 0.0283 [0.0203] [0.0236] [0.0302] [0.0402] Observations 7977 11984 2896 5992 Adj. R-squared 0.654 0.591 0.653 0.612 Matching based on 2005 PM10 in range 21 to 39 All cities Cities > 100,000 Traffic stations Background stations Traffic stations Background stations (1) (2) (3) (4) LEZ treatment -0.0486* 0.0249 -0.0362 0.0431 [0.0272] [0.0264] [0.0346] [0.0415] Observations 10942 17974 5861 9842 Adj. R-squared 0.645 0.581 0.618 0.611

Matching based on 2005 PM10 in range 20 to 40 All cities Cities > 100,000 Traffic stations Background stations Traffic stations Background stations (1) (2) (3) (4) LEZ treatment -0.0486* 0.0228 -0.0362 0.0412 [0.0272] [0.0251] [0.0346] [0.0370] Observations 10942 19686 5861 10698 Adj. R-squared 0.645 0.582 0.618 0.612

Matching based on 2005 PM10 No Range All cities Cities > 100,000 Traffic stations Background stations Traffic stations Background stations (1) (2) (3) (4) LEZ treatment -0.0554** 0.0347 -0.0322 0.0477 [0.0240] [0.0290] [0.0299] [0.0376] Observations 14583 22682 8688 12410 Adj. R-squared 0.669 0.587 0.661 0.618 All regressions include year-month fixed effects, weather, holiday, station type and population covariates. Regressions include data for April-October 2007 vs. 2008. Robust standard errors in brackets are clustered by city, *** p<0.01, ** p<0.05, * p<0.1. 35

Table 10: Effect of Individual LEZ on log PM10 Matching based on 2005 PM10 in range 25 to 35 All station Traffic Background types stations stations (A) Mannheim LEZ (1a) (2a) (3a) LEZ treatment -0.0837*** -0.132*** -0.0345 [0.0182] [0.0187] [0.0263] Observations 11958 5538 6420 Adj. R-squared 0.592 0.618 0.578 (B) Cologne LEZ (1b) (2b) (3b) LEZ treatment -0.00605 -0.0475** 0.0132 [0.0170] [0.0192] [0.0251] Observations 12412 5564 6848 Adj. R-squared 0.601 0.628 0.586 (C) Reutlingen (1c) (2c) (3c) LEZ treatment -0.0461** -0.130*** 0.0342 [0.0206] [0.0197] [0.0271]

Observations 8555 3965 6420 Adj. R-squared 0.632 0.658 0.592 (D) Leonberg (1d) (2d) LEZ treatment -0.0765*** -0.0693*** [0.0194] [0.0187] Observations 11531 5539 Adj. R-squared 0.601 0.631

All regressions include year-month fixed effects, weather, holiday, station type and population covariates. Regressions include data for April-October 2007 vs. 2008. Robust standard errors in brackets are clustered by city, *** p<0.01, ** p<0.05, * p<0.1.

36

Table 11: All Early LEZs vs. Nearby Future LEZs Dirtiest All stations stations Inside LEZ Outside LEZ Traffic stations (1a) (2a) (3a) (4a) LEZ treatment -0.0733** -0.107** -0.0862*** -0.0363 [0.0319] [0.0392] [0.0289] [0.0465] Observations 15794 7849 14156 10885 Adj. R-squared 0.683 0.707 0.692 0.677 Background stations (1b) (2b) (3b) (4b) LEZ treatment 0.0163 0.148 0.0145 0.0178 [0.0381] [0.0549] [0.0643] [0.0327] Observations 9842 1284 6848 8130 Adj. R-squared 0.633 0.556 0.633 0.627 All regressions include year-month fixed effects, weather, holiday, station type and population covariates. Regressions include data for April-October 2007 vs. 2008. Robust standard errors in brackets are clustered by city, *** p<0.01, ** p<0.05, * p<0.1.

Table 12: Placebo Regression: 2005 PM10 Matching Approach – Analogue to Table 8, but using 2007 as the placebo treatment LEZ vs. Attainment cities – 2006 vs. 2007 All cities Cities > 100,000 Traffic Background Traffic Background stations stations stations stations (1) (2) (3) (4) Placebo dummy 0.139 -0.0137 0.121 0.0191 [0.144] [0.0743] [0.0953] [0.0630] Observations 6420 8132 2896 4280 Adj. R-squared 0.670 0.645 0.653 0.653 All regressions include year-month fixed effects, weather, holiday, station type and population covariates. Regressions include data for April-October 2006 vs. 2007. Robust standard errors in brackets are clustered by city, *** p<0.01, ** p<0.05, * p<0.1.

37

Table 13: Placebo Regression of Geographical Matching Approach— Analogue to Table 11, but using 2007 as the placebo treatment All Early LEZs vs. Nearby Future LEZs – 2006 vs. 2007 Dirtiest All stations stations Inside LEZ Outside LEZ Traffic stations (1a) (2a) (3a) (4a) Placebo dummy -0.0256 -0.0066 -0.0189 -0.0528 [0.0405] [0.0764] [0.0470] [0.0674] Observations 17110 7271 15612 11332 Adj. R-squared 0.692 0.700 0.696 0.685 Background stations (1b) (2b) (3b) (4b) Placebo dummy -0.0515 0.0376 0.0034 -0.0845 [0.0443] [0.100] [0.0300] [0.0555] Observations 10690 1707 7694 8555 Adj. R-squared 0.656 0.622 0.650 0.647 All regressions include year-month fixed effects, weather, holiday, station type and population covariates. Regressions include data for April-October 2006 vs. 2007. Robust standard errors in brackets are clustered by city, *** p<0.01, ** p<0.05 * p<0.1

Table 14: Berlin LEZ: Stations within LEZ compared to those outside Stations within LEZ compared to those outside Compared to nearby FLEZ cities Traffic Background Background Traffic stations stations stations stations (1) (2) (3) (4) LEZ treatment -0.0596** -0.0066 [0.0251] [0.0236] LEZ treatment: inside LEZ -0.150** -0.0462 [0.0210] [0.0133] LEZ treatment: outside LEZ -0.0906** -0.0402 [0.0210] [0.0142] Observations 2188 1639 4376 2186 Adjusted R-squared 0.599 0.615 0.591 0.591 All regressions include year-month fixed effects, weather, holiday, station type and population covariates. Regressions include data for February-October 2007 vs. 2008. Robust standard errors in brackets are clustered by month for columns (1) and (2) and city for columns (3) and (4), *** p<0.01, ** p<0.05, * p<0.1.

38

Table 15: Vehicle Registration by emissions sticker category private vehicles (A) Private Vehicles

2008 2009 2010 % change 2008 vs. 2010 Green 34,020,748 34,862,420 35,795,940 5.2 Yellow 3,931,262 3,597,594 3,425,119 -12.9 Red 1,267,825 1,092,315 907,543 -28.4 No Sticker 1,597,089 1,381,064 1,236,204 -22.6 Total 40,816,924 40,933,393 41,364,806 1.3 (B) Commercial Vehicles 2008 2009 2010 % change 2008 vs. 2010 Green 524,542 792,577 985,245 87.8 Yellow 945,181 844,803 793,008 -16.1 Red 469,853 413,133 372,962 -20.6 No Sticker 609,948 518,545 451,169 -26.0 Total 2,549,524 2,569,058 2,602,384 2.1

Table 16: Effect of distance from LEZ city on green technology adoption Change in % of green sticker vehicles from 2008 to 2010 Private vehicles Commercial vehicles (1) (2) Distance to nearest LEZ, km -0.0061*** -0.0108*** [0.0005] [0.0025] Observations 405 405 Adj R-squared 0.265 0.042 Change in % of no sticker vehicles from 2008 to 2010 Private vehicles Commercial vehicles (3) (4) Distance to nearest LEZ, km 0.0030*** 0.0070*** [0.00021] [0.0012] Observations 405 405 Adj R-squared 0.333 0.080 Robust standard errors in brackets, *** p<0.01, ** p<0.05, * p<0.1.

39 Figure 1: Current and Future German LEZs

Figure 2: Classification of cities treatment status

LEZ cities Stations within LEZ (before Oct 2008) Attainment cities (AC) (LEZ) Stations outside Future LEZs (Oct LEZ 2008 and after) Non-attainment (FLEZ) cities = Action Plan cities (AP) Action plan only cities (APO)

40 Figure 3: Average daily PM10 levels by Attainment Status

100 80 60 PM10 40 20 0

01jan2005 01jan2006 01ja n2 007 01jan2008 01jan2009 Note: Each dot represents the national average daily PM10dt2 level. The bold light grey line displays average daily PM10 level for non-attainment cities and the black bold black line the average daily PM10 level for attainment cities both estimated by the locally weighted scatterplot smoothing method with bandwidth of 0.03.

41

Figure 4: Treatment Effects of Individual LEZs Using Regional Approach

Berlin

Stuttgart

Hannover

Cologne

Mannheim

Reutlingen

Tübingen

Ludwigsburg

Leonberg

-.4 -.3 -.2 -.1 0 .1 .2 .3 .4 Coefficients

Traffic stations Background stations

Note: Plots include 95% confidence intervals of treatment effects

42 Figure 5: Change of share of green sticker vehicles 2008 to 2009 as function of distance of the county to LEZ (privately owned cars)

Figure 6: Change of share of no sticker vehicles 2008 to 2009 as function of distance of the county to LEZ (privately owned cars)

43 Figure 7: Change of share of red sticker vehicles 2008 to 2009 as function of distance of the county to LEZ (privately owned cars)

Figure 8: Change of share of yellow sticker vehicles 2008 to 2009 as function of distance of the county to LEZ (privately owned cars)

44 Technical Appendix to

KEEP YOUR CLUNKER IN THE SUBURB: LOW EMISSION ZONES AND ADOPTION OF GREEN VEHICLES

Hendrik Wolff

This is the web based online supplemental Appendix of the main article and consists of the following six sections:

Appendix A: Comparative Results of Recent Urban PM10 Studies Appendix B: Characteristics of German Attainment Cities, Nonattainment Cities and LEZ Appendix C: Average daily PM10 level by LEZ Treatment Status Appendix D: Test of Alternative Specifications: Appendix E: Sample Details on Geographical Matching Approach Appendix F: Cost Benefit Analysis

1

Table A.1: Comparative Results of Recent Urban PM10 Studies

PM 10 Sources

Motor Combustion Study Country Station Vehicle Resuspension (Industry and Natural Type Exhaust (Dust) Individual) Sources Other 1 2 Lenschow et al. Germany Traffic 38% 12% 24% 12% NA (2001) (Berlin) Background 23% 8%3 33% 14%2 NA Querol et al. EU Traffic 35-55% NA 15-25% 17-24% NA (2004) Querol et al. Spain Traffic 54% NA NA 30% 17%4 (2001) 5 Ferusjo et al. Traffic 36% 23% 14% NA 26% Sweden (2007) Background 13% 23% 19% NA 34%5 6 Rodriquez et al. Traffic 25% 33% 16% 11% NA Spain (2003) Background 8% 42% 20% 11%6 NA Chow et al. USA Traffic 30-42% 25-37% NA 18-23%6 NA (1996) (CA) Harrison et al. UK Traffic 32% 50% NA NA 18%7 (1997) 1The authors attribute 50% of PM10 levels to motor vehicles and then split this into 38% from emissions/tire abrasion and 12% from the resuspension of dust caused by traffic. 2The residual is attributed to natural sources such as pollen and wind-borne soil. 3The authors attribute 31% of PM10 levels to traffic and then split this into 23% from emissions/tire abrasion effect on background levels and 8% from resuspension of dust. 4 Source is undetermined. 5Long range transport of pollution or dust particles from outside of Sweden. 6The specific natural source is marine aerosol. 7They identify the residual as secondary ammonium salts and are unable to determine whether these arise from combustion or are the effect of marine air.

2

Appendix B: Characteristics of German Attainment Cities, Nonattainment Cities and LEZ Figure B.1: The LEZ of Stuttgart

Copyright: Landesvermessungsamt Baden-Württemberg, Bundesamt für Kartographie und Geodäsie 2003. The English term “Low Emission Zones” is commonly known in German as Umweltzone (Environmental Zone).

3

Table B1: Current and Future German LEZs Inhabitants Dates of future Future excluded Excluded Size of LEZ: that live City Start date restrictions (2nd, vehicles (2nd, vehicles new within the 3rd round) 3rd round) LEZ Berlin 1/1/2008 no sticker 88 km2 1.1 mill 1/1/2010 red + yellow Cologne 1/1/2008 no sticker 16 km2 130,000 1/1/2010 red Hannover 1/1/2008 no sticker 50 km2 218,000 01/01/09, 01/01/10 red, yellow Dortmund (Brackeler Road) 1/12/2008 no sticker + red < 0.1 km2 300 1/1/2010 not yet planned Ilsfeld 3/1/2008 no sticker 2.5 km2 4,000 1/1/2012 red Leonberg 3/1/2008 no sticker 30 km2 40,000 1/1/2012 red Ludwigsburg 3/1/2008 no sticker 30 km2 55,000 1/1/2012 red Mannheim 3/1/2008 no sticker 7.5 km2 93,900 1/1/2012 red Reutlingen 3/1/2008 no sticker <10 km2 Unknown 1/1/2012 red Schwäbisch Gmünd 3/1/2008 no sticker 5 km2 20,000 1/1/2012 red Stuttgart 3/1/2008 no sticker 207 km2 590,000 1/1/2012 red Tübingen 3/1/2008 no sticker ≈13 km2 Unknown 1/1/2012 red Pleidelsheim 7/1/2008 no sticker 7 km2 7,000 1/1/2012 red Bochum 10/1/2008 no sticker 58.1 km2 150,000 end of 2010 red + yellow Bottrop 10/1/2008 no sticker ≈25 km2 Unknown end of 2010 red + yellow Dortmund 10/1/2008 no sticker 19.1 km2 587,137 1/1/2011 red Duisburg 10/1/2008 no sticker ≈43 km2 Unknown end of 2010 red + yellow Essen 10/1/2008 no sticker 140 km2 14,000 1/1/2011 red Frankfurt 10/1/2008 no sticker 110 km2 Unknown 01/01/10, 01/01/12 red, yellow Gelsenkirchen 10/1/2008 no sticker 20 km2 Unknown end of 2010 red + yellow Mülheim 10/1/2008 no sticker ≈14.2 km2 Unknown end of 2010 red + yellow München 10/1/2008 no sticker 44 km2 431,000 1/1/2010 red Oberhausen 10/1/2008 no sticker 23.8 km2 91,000 end of 2010 red + yellow Recklinghausen 10/1/2008 no sticker <20 km2 Unknown Unknown Unknown Bremen 1/1/2009 no sticker 7 km2 56,000 1/1/2010 red Heilbronn 1/1/2009 no sticker ≈22.5 km2 Unknown 1/1/2012 red Herrenberg 1/1/2009 no sticker ≈4 km2 28,000 1/1/2012 red Karlsruhe 1/1/2009 no sticker ≈ 12 km2 Unknown 1/1/2012 red Mühlacker 1/1/2009 no sticker ≈ 1.5 km2 Unknown 2012 red Pforzheim 1/1/2009 no sticker ≈ 2 km2 Unknown 1/1/2012 red Ulm 1/1/2009 no sticker ≈ 27 km2 Unknown 1/1/2012 red Düsseldorf 2/15/2009 no sticker 13.8 km2 36,500 1/1/2011 red Wuppertal 2/15/2009 no sticker ≈ 15 km2 Unknown 1/1/2011 red Augsburg 7/1/2009 no sticker 5.2 km2 Unknown 1/1/2010 red Neu-Ulm 11/1/2009 no sticker ≈ 2.7 km2 Unknown 1/1/2012 red Bonn 1/1/2010 no sticker ≈ 5 km2 Unknown 7/1/2011 red + yellow Freiburg 1/1/2010 no sticker 28 km2 120,000 1/1/2012 red + yellow Heidelberg 1/1/2010 no sticker 10.3 km2 170,000 1/1/2012 red Münster 1/1/2010 no sticker + red ≈ 1.5 km2 Unknown Unknown Unknown Osnabrück 1/4/2010 no sticker 14 km2 7,000 1/4/2011 red Pfinztal 1/1/2010 no sticker 31 km2 18,000 1/1/2012 red Dresden 2012 no sticker 4.2 km2 6,500 Unknown Unknown Leipzig 1/1/2011 no sticker-yellow ≈ 239 km2 Unknown Unknown Unknown Cities with proposed LEZs Arnsbach, Arzberg, Aschersleben, Bayreuth, Bernau, Brandenburg an der Havel, Braunschweig, (no final start date yet): Burgdorf, Burghausen, Castrop-Rauxel, Chemnitz, Cottbus, Darmstadt, Eberswalde, Erfurt, Erwitte, Frankfurt an der Oder, Gera, Görlitz, Halle (Saale), Hambach, Hamburg, Ingolstadt, Itzehoe, Jena, Kassel, Krefeld, Lahn-Dill, Landshut, Lindau, Ludwigshafen, Lutherstadt Wittenberg, Mainz, Magdeburg, Mühlheim an der Ruhr, Nauen, Neuruppin, Neuwied, Neuss, Passau, Potsdam, Regensburg, Rhein-Main, Schwandorf, Speyer, Trier, Warstein, Weiden, Weimar, Worms, Wuppertal, Würzburg 4

Table B2: Characteristics of all Attainment and Nonattainment Cities 2005 Violate Avg. 2005 PM 10 Exceed- limit in LEZ start City at highest Treatment status Population ance 2005- date polluting station days 06 Wascheid 12.0 0 0 Attainment Netphen 12.6 0 0 Attainment Neuglobsow 13.8 3 0 Attainment Simmerath 14.0 0 0 Attainment Welzheim 16.2 4 0 Attainment Andechs, Gde.teil Rothenfeld 16.5 4 0 Attainment 3,237 Dunzweiler 16.6 2 0 Attainment 974 Hummelshain 16.6 1 0 Attainment 641 Bad Arolsen/Kohlgrund 17.0 5 0 Attainment Wittenberge 17.3 2 0 Attainment Dreißigacker 17.5 0 0 Attainment Rehlingen-Siersburg 17.7 3 0 Attainment 15,805 Klötze 17.8 2 0 Attainment 5,243 Kiel 18.7 5 0 APO-no violation 234,470 Güstrow 19.4 4 0 Attainment 105,071 Saarlouis 19.6 3 0 Attainment 209,719 Westerland 19.6 7 0 Attainment Kempten (Allgäu) 19.7 8 0 Attainment 61,442 Pfullendorf 20.1 8 0 Attainment Soest 20.4 6 0 Attainment 308,211 Wörth 20.5 8 0 Attainment 17,500 Tauberbischofsheim 20.5 13 0 Attainment Gülzow 20.6 9 0 Attainment 1,288 Wilhelmshaven 20.8 11 0 Attainment 83,245 Ratingen 20.8 5 0 Attainment Leverkusen 20.8 2 0 Attainment 161,030 Zarrentin 20.8 9 0 Attainment 4,672 Solingen 20.9 7 0 Attainment 163,291 Naila 21.1 7 0 Attainment 8,305 Walsrode 21.1 8 0 Attainment Michelstadt 21.2 7 0 Attainment Zella-Mehlis 21.3 4 0 Attainment 12,245 Göhlen 21.3 11 0 Attainment 407 Tübingen 21.6 9 1 LEZ 3/1/2008 216,616 Biberach 21.6 13 0 Attainment 188,693 Klingenthal 21.6 9 0 Attainment 8,831 Pforzheim 21.7 13 1 'Future' LEZ 1/1/2009 119,168 Eisenach 21.8 10 0 Attainment 43,703 Jork 21.8 11 0 Attainment Völklingen 21.9 3 0 Attainment 40,794 Nettetal 22.1 8 0 Attainment Reidstadt 22.2 9 0 Attainment Eggenstein 22.3 10 0 Attainment Neuruppin 22.4 13 0 APO-no violation Wiesloch 22.4 12 0 Attainment 5

2005 Violate Appendix B (cont.) Avg. 2005 PM 10 Exceed- limit in LEZ start at highest Treatment status Population ance 2005- date City polluting station days 06 Dillingen 22.5 4 0 Attainment 21,431 Friedrichshafen 22.5 14 0 Attainment Kleinwallstadt 22.6 9 0 Attainment 5,823 Fulda 22.7 7 0 Attainment 219,600 Neu Zauche 22.7 16 0 Attainment Aalen 22.8 16 0 Attainment Bonn 22.9 4 0 'Future' LEZ -no violation 1/1/2010 313,291 Raunheim 23.1 12 0 Attainment Zeitz 23.1 16 0 Attainment 31,045 Hattingen 23.2 7 0 Attainment Wesel 23.2 15 0 Attainment 475,923 Radebeul 23.2 14 0 Attainment 33,091 Greiz 23.2 16 0 Attainment 115,387 Waiblingen 23.3 13 0 Attainment Bebra 23.3 10 0 Attainment Neustadt a.d. Donau 23.3 14 0 Attainment 12,738 Schwerte 23.5 9 0 Attainment Lünen 23.5 11 0 Attainment Osnabrück 23.6 13 1 'Future' LEZ -no violation 1/4/2010 163,330 Konstanz 23.6 18 0 Attainment 274,571 Plochingen 23.6 13 0 Attainment Delitzsch 23.7 12 0 Attainment 122,500 Buckow 23.8 21 0 Attainment Schwäbisch Hall 23.9 13 0 Attainment 189,579 Saalfeld 24.0 16 0 Attainment 27,861 Heidelberg 24.0 11 0 'Future' LEZ -no violation 1/1/2010 143,897 Burg 24.0 6 0 Attainment 25,000 Lingen 24.4 21 0 Attainment Meiningen 24.4 10 0 Attainment 21,448 Hof 24.4 21 0 Attainment 48,443 Hoyerswerda 24.4 20 0 Attainment 42,048 Bernburg 24.4 9 0 Attainment 64,860 Rostock 24.7 15 0 APO-no violation 199,325 Zwickau 24.7 18 0 Attainment 97,296 Hürth 24.7 8 0 Attainment Suhl 24.8 2 0 Attainment 42,283 Speyer 24.8 18 0 APO-no violation 50,567 Kulmbach 24.9 12 0 Attainment 76,890 Mönchengladbach 25.0 24 0 Attainment 261,216 Ulm 25.1 18 1 'Future' LEZ 1/1/2009 120,748 Schweinfurt 25.1 14 0 Attainment 54,097 Altenburg 25.2 27 0 Attainment 37,236 Coburg 25.4 15 0 Attainment 41,768 Aschaffenburg 25.6 12 0 Attainment 68,645 Wiesbaden 25.8 18 0 Attainment 275,085 Bernhausen 25.9 21 1 APO 13,216 6

2005 Violate Appendix B (cont.) Avg. 2005 PM 10 Exceed- limit in LEZ start at highest Treatment status Population ance 2005- date City polluting station days 06 Bautzen 25.9 20 0 Attainment 148,945 Stralsund 26.2 22 0 Attainment 58,563 Heilbronn 26.2 22 1 'Future' LEZ 1/1/2009 121,498 Lindau (Bodensee) 26.3 28 1 APO 79,636 Emden 26.3 20 0 Attainment 51,666 Nauen 26.4 25 0 APO-no violation 16,674 Hanau 26.5 20 0 Attainment 88,251 Königs Wusterhausen 26.5 20 0 Attainment 33,201 Weißenfels 26.6 32 0 Attainment 73,624 Pirmasens 26.6 16 0 Attainment 42,761 Bamberg 26.7 20 0 Attainment 69,746 Freiberg 26.7 33 0 Attainment 144,094 Leonberg 26.8 16 1 LEZ 3/1/2008 45,537 Stendal 26.9 18 0 Attainment 130,436 Gelsenkirchen 27.0 24 0 'Future' LEZ 10/1/2008 267,418 Cologne 27.0 14 0 LEZ-no violation 1/1/2008 986,317 Mülheim 27.0 21 0 'Future' LEZ-no violation 10/1/2008 169,651 Zittau 27.0 31 0 Attainment 29,898 Arzberg 27.0 24 0 APO-no violation 5,893 Itzehoe 27.1 21 0 APO-no violation 33,800 Dessau 27.2 18 0 Attainment 77,914 Schwandorf 27.3 30 0 APO-no violation 144,644 Worms 27.5 27 1 APO 81,984 Würzburg 27.7 30 0 APO-no violation 134,080 Glauchau 27.8 24 0 Attainment 25,760 Norderney 27.8 17 0 Attainment 5,986 Aachen 28.0 18 0 APO 258,055 Wuppertal 28.0 20 0 'Future' LEZ 2/15/2009 358,813 Plauen 28.1 33 1 APO 68,614 Magdeburg 28.3 22 1 APO 229,344 Erlangen 28.3 22 0 APO 103,469 Gera 28.4 31 1 APO 103,446 Reutlingen 28.5 17 1 LEZ 3/1/2008 281,933 Saarbrücken 28.5 18 0 Attainment 340,702 Ratzeburg 28.8 28 0 APO-no violation 13,671 Datteln 29.0 30 0 Attainment 36,297 Krefeld 29.0 24 1 APO 237,336 Borna 29.1 31 0 Attainment 22,561 Neu-Ulm 29.1 34 1 'Future' LEZ 11/1/2009 163,477 Jena 29.6 29 1 APO 102,291 Landshut 29.7 39 1 APO 61,757 Nürnberg 29.7 33 0 APO-no violation 498,936 Weimar 29.8 35 1 APO 64,541 Trier 29.9 26 0 APO-no violation 100,198 Karlsruhe 29.9 22 1 'Future' LEZ 1/1/2009 285,756 Bottrop 30.0 33 0 'Future' LEZ 10/1/2008 119,195 7

2005 Violate Appendix B (cont.) Avg. 2005 PM 10 Exceed- limit in LEZ start at highest Treatment status Population ance 2005- date City polluting station days 06 Fürth 30.1 30 0 Attainment 113,596 Ansbach 30.2 29 1 APO 40,531 Regensburg 31.6 37 1 APO 130,153 Ludwigshafen 31.7 37 1 APO 163,536 Görlitz 31.8 42 1 APO 57,418 Hagen 32.0 27 1 APO 196,295 Halle/Saale 32.2 51 1 APO 236,576 Kassel 32.2 48 1 APO 193,842 Aschersleben 32.2 38 1 APO 31,717 Freiburg 32.5 21 1 'Future' LEZ 1/1/2010 216,448 Münster 32.5 33 0 'Future' LEZ-no violation 1/1/2010 271,404 Frankfurt 32.5 48 1 'Future' LEZ 10/1/2008 648,925 Mannheim 33.4 43 1 LEZ 3/1/2008 307,847 Mainz 33.7 47 1 APO 195,178 Hamburg 33.7 46 1 APO 1,748,544 Darmstadt 34.0 42 1 APO 140,366 Erfurt 34.3 49 1 APO 202,723 Bayreuth 34.9 54 1 APO 73,617 Dresden 34.9 78 1 'Future' LEZ 2011 500,471 Potsdam 35.2 55 1 APO 148,126 Pleidelsheim 35.6 55 1 LEZ 7/1/2008 6,239 Essen 35.9 61 1 'Future' LEZ 10/1/2008 584,136 Frankfurt (Oder) 36.9 65 1 APO 63,177 Augsburg 37.1 61 1 'Future' LEZ 7/1/2009 262,492 Hannover 37.5 63 1 LEZ 1/1/2008 515,559 Düsseldorf 38.0 69 1 'Future' LEZ 2/15/2009 576,090 Berlin 38.1 74 1 LEZ 1/1/2008 3,399,896 Leipzig 38.2 75 1 'Future' LEZ 1/1/2011 504,798 Dortmund 39.5 82 1 'Future' LEZ 10/1/2008 587,870 Duisburg 40.0 83 1 'Future' LEZ 10/1/2008 500,217 Ludwigsburg 41.1 78 1 LEZ 3/1/2008 513,799 München 44.8 107 1 'Future' LEZ 10/1/2008 1,278,559 Stuttgart 54.5 187 1 LEZ 3/1/2008 593,244 Berghausen NA NA 1 APO Bernau NA NA 0 APO-no violation Burgdorf NA NA 0 APO-no violation Edertal-Hemfurth NA NA 0 Attainment Flensburg NA NA 0 Attainment 86,365 Heidenheim NA NA 0 Attainment 134,722 Heppenheim NA NA 0 Attainment Herrenberg NA NA 1 'Future' LEZ 1/1/2009 Ilsfeld NA NA 1 LEZ 3/1/2008 8,307 Markgröningen NA NA 0 Attainment Mühlacker NA NA 1 'Future' LEZ 1/1/2009 Possen NA NA 0 Attainment Sproitz NA NA 0 Attainment 8

2005 Violate Appendix B (cont.) Avg. 2005 PM 10 Exceed- limit in LEZ start at highest Treatment status Population ance 2005- date City polluting station days 06 Wlzbachtal-Jöhlingen NA NA 0 Attainment Notes: Shaded area used in PM10 matching analysis. List only includes stations with sufficient data. 'Future' LEZs came into effect on or after 10/1/2008. 'No violation' refers to cities with APs despite not violating the PM10 standard.

9

Appendix C Figure: Average daily PM10 level by LEZ Treatment Status Panel a: 2005 PM10 Matching Approach 100 80 60 PM10 40 20 0

01jan2005 01jan2006 01jan2007 01jan2008 01jan2009

Panel b: Geographical Matching Approach 100 80 60 PM10 40 20 0

01jan2005 01jan2006 01jan2007 01jan2008 01jan2009

Note: Each dot represents the average daily PM10 level of the samples described under each of the two approaches. (The sample of the 2005 matching approach is described in Section 4.2 and the sample of the Geographical approach is described in Section 4.3). The bold light grey line displays average daily PM10 level for control cities and the black bold black line the average daily PM10 level for treatment cities both estimated by the locally weighted scatterplot smoothing method with bandwidth of 0.04.

10

Appendix D: Test of Alternative Specifications: With respect to robustness in covariates, the table below lists the effects of including/omitting the following set of regressors:

o Original regression including all covariates o Without any weather covariates o Without Holiday covariates o Without Population covariates o Without any covariate, except the necessary dummies to identify the Differences-in-Differences treatment effects, Table: LEZ vs. Attainment cities – All Cities Matching based on 2005 PM10 in range 25 to 35 With All Covariates Without Weather Covariates Without Holiday Covariates Background Background Background Traffic stations stations Traffic stations stations Traffic stations stations (1) (2) (3) (4) (5) (6) LEZ treatment -0.0910*** 0.00724 -0.105*** 0.0100 -0.0912*** 0.00722 [0.0241] [0.0285] [0.0244] [0.209] [0.0247] [0.0287] Observations 6723 7704 6723 7704 6723 7704 Adj. R-squared 0.657 0.591 0.314 0.197 0.649 0.558 LEZ vs. Attainment cities – Cities > 100,000 Matching based on 2005 PM10 in range 25 to 35 With All Covariates Without Weather Covariates Without Holiday Covariates Background Background Background Traffic stations stations Traffic stations stations Traffic stations stations (1) (2) (3) (4) (5) (6) LEZ treatment -0.0686* 0.0448 -0.0663* 0.0559* -0.0685* 0.0454 [0.0302] [0.0354] [0.0307] [0.0265] [0.0310] [0.0357] Observations 2896 4280 2896 4280 2896 4280 Adj. R-squared 0.653 0.612 0.300 0.193 0.641 0.608

11

Table: LEZ vs. Attainment cities – All Cities Matching based on 2005 PM10 in range 25 to 35 With All Covariates Without Population Covariates Without Any Covariates Background Background Background Traffic stations stations Traffic stations stations Traffic stations stations (1) (2) (3) (4) (5) (6) LEZ treatment -0.0910*** 0.00724 -0.0910*** 0.00724 -0.106*** 0.0102 [0.0241] [0.0285] [0.0241] [0.0285] [0.0248] [0.209] Observations 6723 7704 6723 7704 6723 7704 Adj. R-squared 0.657 0.591 0.657 0.591 0.299 0.187 LEZ vs. Attainment cities – Cities > 100,000 Matching based on 2005 PM10 in range 25 to 35 With All Covariates Without Population Covariates Without Any Covariates Background Background Background Traffic stations stations Traffic stations stations Traffic stations stations (1) (2) (3) (4) (5) (6) LEZ treatment -0.0686* 0.0448 -0.0686* 0.0448 -0.0669* 0.0564* [0.0302] [0.0354] [0.0302] [0.0354] [0.0313] [0.0265] Observations 2896 4280 2896 4280 2896 4280 Adj. R-squared 0.653 0.612 0.653 0.612 0.283 0.181 Except where indicated in the column header, all regressions include year-month fixed effects, weather, holiday, station type and population covariates. Regressions include data for April-October 2007 vs. 2008. Robust standard errors in brackets are clustered by city, *** p<0.01, ** p<0.05, * p<0.1.

These alternative specifications of Table 8 show that our results are overall qualitatively similar to those when all covariates are included.

12

Appendix E: Sample Details on Geographical Matching Approach For the regional regressions, the following control cities are used for each LEZ city:

Table E1: Control Cities for Individual LEZ Regressions

Stuttgart, Tübingen, Reutlingen & Ludwigsburg Leonberg Mannheim Cologne Hannover Berlin Heidelberg Herrenberg Heidelberg Essen Bremen Leipzig Karlsruhe Mühlacker Karlsruhe Dortmund Osnabruck Dresden Pforzheim Dusseldorf Göttingen Ulm Duisburg Braunschweig Heilbronn Freiburg Herrenberg Mühlacker

13

Table E2: Effect of individual LEZs on log PM10 Matching based on regional approach Berlin Stuttgart Hannover Cologne Mannheim Reutlingen Tubingen Ludwigsburg Leonberg Traffic stations (1a) (2a) (3a) (4a) (5a) (6a) (7a) (8a) (9a) LEZ treatment -0.120*** -0.0288 -0.0939** -0.0742 -0.0992 -0.0582** -0.0296 0.0489* 0.0687 [0.0352] [0.0218] [0.0215] [0.0416] [0.0553] [0.0246] [0.0213] [0.0212] [0.0819] Observations 4376 6507 2188 2996 2050 4836 4879 4880 1202 Adj. R-squared 0.59 0.712 0.579 0.685 0.633 0.667 0.647 0.668 0.436 Background stations (1b) (2b) (3b) (4b) (5b) (6b) (7b) (8b) (9b) LEZ treatment -0.0442 0.262*** 0.0516 -0.0837 0.114* 0.118** 0.159*** 0.0217 [0.0494] [0.0243] [0.0330] [0.0425] [0.00894] [0.0244] [0.0240] [0.0370] Observations 2186 1712 2735 2568 856 1712 1712 1712 Adj. R-squared 0.591 0.619 0.461 0.612 0.639 0.596 0.591 0.593 All regressions include year-month fixed effects, weather, holiday, station type and population covariates. Robust standard errors in brackets are clustered by city, *** p<0.01, ** p<0.05, * p<0.1

14 Appendix F: Cost Benefit Analysis F.1 Benefits

We use improvements in long-term mortality attributable to the decreased PM10 in LEZs as our measure of benefits. Long-term mortality measures the decrease in life expectancy caused

by long-term exposure to PM10. We ignore acute mortality, or the increase in mortality due to a

short-term increase in PM10, since this may just be measuring the ‘harvesting’ effect where

people who were near death die a few days or weeks earlier. To calculate the effect of PM10 on

long-term mortality, we use estimates of the link between PM10 and mortality and morbidity in France, Switzerland and Austria. These estimates were derived by the World Health Organization (WHO) and have been used extensively in the epidemiology literature, i.e. in Medina et al. (2004), Kunzli et al. (2000), Seethaler (1999), van Zelm (2008).1 Specifically, the WHO study found that for every one million residents in Switzerland and France, each 10 μg/m³

increase in PM10 is associated with an additional 340 premature mortalities. Since these studies

find that the effect of PM10 on mortality is close to linear over the relevant range of PM10, this

means that each 1 μg/m³ increase in PM10 is associated with 34 deaths per million residents. From these numbers using procedure described in section 7, we calculate the number of lives saved by each LEZ using the number of inhabitants within each LEZ. We multiply this by the EPA’s value of statistical life (VSL) of $7,900,000 (2008$)2 to monetize these benefits (EPA 2000). Using this method, as summarized in Table A1 we find that the benefit from LEZs is approximately $1.98 billion ($1,978,395,825).

1 These estimates are based on two cohort studies, Pope, et al. (1995) and Dockery, et al. (1993), as re-estimated by Krewski, et al. (2000). In their extensive review of the literature, the EPA singled out these two as the best studies for their cost-benefit analysis of the Clean Air Act Amendments (EPA 1999). 2 This value has been adjusted to 2008 dollars from the value for 1999 specified in the cited report. Kiesner et al. (2012) estimate a range of VSL from 7 to 12 million.

15 Table F1: Value of mortality benefits from decreased PM10

Fixed baseline mortality increment per 10 μg/m³ PM10 and 1 million inhabitants 340 Deaths per person per 1 μg/m³ 0.000034 Amount Traffic Avg 2007 PM10 station Traffic station decreases in Inhabitants Number of City coefficient PM10 2008 of LEZ lives saved Berlin -0.15003 28.86 4.33 1,300,000 191.33 Ludwigsburg 0.0489 34.65 -1.69 55,000 -3.17 Tubingen -0.0296 31.26 0.93 78300 2.46 Reutlingen -0.0582 38.12 2.22 78523.2 5.92 Stuttgart -0.0288 33.01 0.95 590,000 19.07 Hannover -0.0939 26.02 2.44 218,000 18.11 Leonberg 0.0687 33.42 -2.30 40,000 -3.12 Koln -0.0742 32.98 2.45 130,000 10.82 Mannheim -0.0992 28.43 2.82 93,900 9.00 Total number of lives saved 250

EPA Estimate Value of statistical life $7,800,000 Value of lives saved $1,953,352,840

This estimate of benefits is conservative for many reasons. First, we only count the improvement in mortality amongst people who reside within the LEZs studied. As our results

show, however, PM10 also decreased in traffic areas outside of LEZs, most likely because of the adoption of cleaner vehicles, so if these areas were also included the number of lives saved would be higher. If each city’s entire population was used instead of just inhabitants of the LEZ, the benefits would jump to $5.22 billion ($5,217,522,677). The second way in which our estimates are conservative is that we only consider long-

term mortality. PM10 is also associated with non-lethal morbidity, however. In the above studies, also health effects from respiratory hospital admissions, cardiovascular hospital admissions, adult chronic bronchitis, child bronchitis and adult and child asthmatic attacks are considered. If these conditions and parameters are included in our benefits calculation in the same manner as above,4 then Table A2 shows how our measure of the benefits increases by $13,661,332. F.2 Costs

3 This estimate is derived from the stations that reside inside of the LEZ of Berlin (column 3 of Table 14). 4 For the conditions that differentiate between adults and children, we adjust the population numbers, using 14% as the proportion of children under 14 in Germany. http://www.countryreports.org/people/ageStructure.aspx?countryid=91&countryname=

16 To measure the costs LEZs have imposed on Germans, we estimate the total cost of upgrading vehicles to be able to enter the LEZs. Since we measure the health benefits realized between 2007 and 2008, we also look at the costs of upgrading vehicles over this time period. To do this, we use our spatial vehicle registration data to fit regressions of the change in share of green-sticker cars and trucks from 2008 to 2009 on distance from an LEZ. Since we don’t want to count vehicles that would have switched to green sticker vehicles in the absence of the LEZ regulation, we use the change in share of green stickers for the point furthest away from an LEZ (0.0110 and 0.0828 at 244 km from an LEZ for cars and trucks, respectively) as the baseline change in share of green stickers. For each location, we subtract this 0.0110 (0.0828) from our regression’s predicted change in share of green stickered cars (trucks). This is the change in share of green stickers due to the LEZ, which we then multiply by the number of cars (trucks) for that location in 2008 to get the number of new green cars attributable to LEZs. We sum these numbers for all locations to get the total number of new cars and trucks due to the LEZ and multiply this by the average cost for upgrading a vehicle ($1,650 for cars, $14,500 for trucks) to get the total cost of upgrading cars and trucks because of LEZs. In other words, we estimate cost using the following formula

= 퐽 ( ) , 푇표푡푎푙 퐶표푠푡 � 푝푖 � 푁푖푗 퐶�횤횥 − 퐶푖0 where i represents cars and trucks, j indexes푖=푐푎푟푠 counties,푡푟푢푐푘푠 푗N= 1is the number of vehicles in 2008, is fitted value of change in share of green cars, is the baseline change in share of green vehicles,퐶�횤횥 and p equals the cost of upgrading each vehicle퐶푖0 type. We find that the total cost of upgrading cars is $475,185,312 and the total cost of upgrading trucks is $618,133,842. The combined total cost is $1,093,319,154. This cost is nearly half of our primary measure of benefits, $1,978,395,825. If one considers the benefits for those who live close to but outside of an LEZ, as well as morbidity benefits, then the benefits of LEZs will exceed the costs by even more.

17 Table F2: Value of morbidity benefits from decreased PM10

Fixed baseline mortality Willingness to increment per 10 μg/m³ Deaths per Pay to avoid PM10 and 1 million person per 1 condition Condition inhabitants cases μg/m³ (1996 Euros) Respiratory Hospital Admission 140 0.000014 $7,870.00 Cardiovascular Hospital Admissions 255 0.0000255 $7,870.00 Chronic Bronchitis Incidence (adult) 410 0.000041 $20,900.00 Bronchitis (child) 4725 0.0004725 $131.00 Asthmatic Attacks (children) 2500 0.00025 $31.00 Asthmatic Attacks (adult) 6280 0.000628 $31.00 Number of incidents avoided Chronic Respiratory Cardiovascular Asthmatic Asthmatic Bronchitis Bronchitis Traffic Avg 2007 Amount PM Hospital Hospital Attacks Attacks 10 Incidence (child) station Traffic decreases in Inhabitants Admission Admissions (adult) (children) (adult) City coefficient station PM10 2008 of LEZ Berlin -0.15 28.86 4.33 1,300,000 78.78 143.50 198.42 372.25 3039.26 196.96 Ludwigsburg 0.0489 34.65 -1.69 55,000 -1.30 -2.38 -3.29 -6.17 -50.34 -3.26 Tubingen -0.0296 31.26 0.93 78300 1.01 1.85 2.55 4.79 39.13 2.54 Reutlingen -0.0582 38.12 2.22 78523 2.44 4.44 6.14 11.52 94.09 6.10 Stuttgart -0.0288 33.01 0.95 590,000 7.85 14.30 19.78 37.10 302.92 19.63 Hannover -0.0939 26.02 2.44 218,000 7.46 13.58 18.78 35.24 287.69 18.64 Leonberg 0.0687 33.42 -2.30 40,000 -1.29 -2.34 -3.24 -6.07 -49.60 -3.21 Koln -0.0742 32.98 2.45 130,000 4.45 8.11 11.22 21.05 171.83 11.14 Mannheim -0.0992 28.43 2.82 93,900 3.71 6.75 9.34 17.52 143.03 9.27 Total incidents avoided 103.12 187.82 259.71 487.23 3978.00 257.80 Willingness to pay (1996 Euros) $811,540 $1,478,162 $5,427,949 $63,828 $123,318 $7,992

Willingness to pay total (1996 Euros) $7,912,789 Willingness to pay total (2008 USD) $13,661,332

18 References of the Appendix

Chow, J.C., Watson, J.G., Lowenthal, D.H., and Countess, R.J. (1996). ‘Sources and chemistry of PM10 aerosol in Santa Barbara County, CA’ Atmospheric Environment, vol. 30, pp.1489–1499. Furusjö, E., Sternbeck, J. and Cousins, A.P. (2007). ‘PM (10) source characterization at urban and highway roadside locations.’ The Science of the Total Environment, 387, pp. 206. Harrison, R.M., Deacon, A.R., Jones, M.R. and Appleby, R.S. (1997). ‘Sources and processes affecting concentrations of PM10 and PM2.5 particulate matter in Birmingham (UK)’ Atmospheric Environment, vol. 31, pp. 4103–4117. Kiesner, Viscusi, Woock and Ziliak (2012): The Value of a Statistical Life: Evidence from Panel Data. Review of Economics and Statistics. Vol. 94(1): 74–87. Lenschow, P., Abraham, H.J., Kutzner, K., Lutz, M., Preuß, J.D. and Reichenbächer, W. (2001). ‘Some ideas about the sources of PM10’ Atmospheric Environment, vol. 35, pp. S23–S33. Querol, X., Alastuey, A., Rodriguez, S., Plana, F., Ruiz, C.R., Cots, N., Massagué, G. and Puig, O. (2001). ‘PM10 and PM2. 5 source apportionment in the Barcelona Metropolitan area, Catalonia, Spain.’ Atmospheric Environment, vol. 35, pp. 6407–6419. Querol, X., Alastuey, A., Ruiz, C.R., Artiñano, B., Hansson, H.C., Harrison, R.M., Buringh, E., Ten Brink, H.M., Lutz, M. and Bruckmann, P. (2004). ‘Speciation and origin of PM10 and PM2. 5 in selected European cities.’ Atmospheric Environment, vol. 38, pp. 6547–6555. Rodrı́guez, S., Querol, X., Alastuey, A., Viana, M.M., Alarcón, M., Mantilla, E. and Ruiz, C.R. (2004). ‘Comparative PM10–PM2. 5 source contribution study at rural, urban and industrial sites during PM episodes in Eastern Spain.’ Science of the Total Environment vol. 328, pp. 95–113.

19